The limited partners who provide the capital to venture firms — the “money behind the money” — are feeling great right now. They’re in uncharted territory in many ways, but also potentially exceedingly lucrative territory.

Normally, LPs typically spread their bets across many firms to diversify their own portfolios. They write checks to emerging managers and established ones, to generalist firms and specialists. The goal is a strong return, but also to minimize risk.

That’s the typical playbook. But typical is out the window. OpenAI and Anthropic’s valuations have shot up so quickly that nearly everyone has piled into the same two companies on the way up, often inside the same venture firm, despite the longstanding venture convention of avoiding investments that conflict with one another. No one is complaining, either. The LPs I speak with seem unbothered by this big gamble, given the potential returns; they’re expecting a bonanza.

But they do see danger ahead, even if they describe this as a “yellow light” situation, and they’re uneasy about how quickly VCs are coming back to raise new funds.

I talked about all of it recently with Beezer Clarkson of LGT Capital Partners, a Zurich-based diversified asset manager, and Amit Bhatti of TrueBridge, a Chapel Hill firm that focuses solely on venture. This conversation has been edited for length.

These are strange days — exciting but also scary. How are you feeling about your roles right now, and your place in this startup ecosystem?

Beezer Clarkson: It’s very busy, but I wouldn’t say it’s scary. There’s a lot of change and a lot of variables at play. I love venture, so I find the chaos fun. I’ve been doing this for a couple of decades, so seeing the cycles makes it less scary. You appreciate that these things happen, and that we should enjoy the upside, because sometimes there’s downside, and the up is a lot more fun than the down.

Amit Bhatti: If anything, it’s exciting. Venture is sometimes looked at as a niche, and now that AI is all-consuming, all the attention is on venture. This feels like our moment more than it is like, “Oh gosh, what’s going to happen?” The pace of investing and the valuations may be different, but really meaningful stuff is happening, and all the attention is focused on this corner of private equity.

Your funds all hold the same brands now, OpenAI and Anthropic, multiplied across however many funds you back, and your job is to keep your institutions diversified. How do you think about that? Obviously it’s great right now, but if things turned, it would be less great.

Amit Bhatti: Venture has always been a power law business, so I actually view the concentration as a good thing. If we were getting to the maturity of one of our funds and there were still lots of tiny positions, something’s gone wrong. Our funds that are 10-plus years old are starting to consolidate into a few names. There’s a lot of Databricks, a lot of SpaceX, a lot of Stripe. In this generation of funds, I would hope there’s a lot of Anthropic and a lot of OpenAI. That’s a feature, not a bug. Not every one of these companies is going to work out, but that consolidation is typical. What’s different is how quickly it’s happened. I don’t view it as an immediate panic signal.

Beezer Clarkson: At the early stage, you still see quite a diversity of opportunity sets, and the dollars are smaller. It aggregates up into the same companies, but early-stage funds obviously aren’t doing the big late-stage stuff, so you have more diversity there, if that’s what you’re looking for.

Funds that invested on four-year cycles have shrunk that to three, then two, then 18 months. Is the pacing okay, or is it concerning?

Beezer Clarkson: We’re seeing funds come back on one-year pacing, which is fast. The norm was probably two to two and a half years. Some funds are deploying much faster because they’re taking concentrated, big-check positions. That could be a worrisome sign. It could work out, we don’t know, but it’s typically harder to do a one-year vintage and get it right.

Amit Bhatti: Similar for us. I would have said 24 to 36 months was roughly normal. Now 12 to 18 is not crazy for some of the funds going really quickly. A lot of funds are bumping up their fund sizes mid-cycle, which is definitely unusual. That’s partly because rounds have gotten so large. If Anthropic is raising a giant quantum of capital and you put 20% of the fund into it, you’re obviously going to be ahead of pace. Seed round sizes have gone up almost 2x, too, so if you want to lead, you’re writing 2x the check. It’s definitely a yellow light. It gives us a little bit of heartburn.

But when we look at what these managers actually bought, the numbers are different. The last time we had accelerated pacing, there was a lot of investing in zero-revenue crypto moonshots. Now we’re seeing a lot of companies go from zero to $100 million in revenue, which was totally unheard of before.

If you’ve got a great manager who comes back every year but the returns are still on paper, can you say “call me next time,” or do you lose your seat at the table?

Beezer Clarkson: It’s very hard to leave a manager and then come back. It happens, but it’s more typical to downsize your check. In 2021, a lot of LPs said, “If I was going to give you $10 million, I’ll do $5 million, because $10 million was for two years.” In our portfolios, the pacing gets pulled through, so we think about our pacing as well as yours.

Amit Bhatti: Same here. We’re building out a calendar of when we expect to make commitments, so if someone goes way off that map, we need to make sure the capital is there. You see people downsize or communicate it up front. It’s hard to say you’re sitting it out altogether, because then you’re not reliable capital. That’s also why you’re seeing upsizes; managers don’t want to come back and say they’re raising a whole new fund, so they bump up the old one.

Beezer Clarkson: People also rotate out of venture sometimes because capital calls keep running with no distributions coming back, and they still have to meet operating commitments, whether it’s the endowment, the hospital, or the foundation. That puts a lot of strain on those situations.

LPs like time diversification, yet a lot of funds were raised in recent years, including that backed startups at high valuations during the pandemic. How are those funds doing, and how would you describe 2026 so far?

Beezer Clarkson: The 2021 vintages are still early, five years in, but they argue against doing one-year funds. Some will catch a piece of an AI company and look excellent, but your odds of getting it right in one year are just tougher, and the valuations were extreme.

Amit Bhatti: If I look at the last 15 years of our portfolio, I wouldn’t bet on the COVID vintage being the best one. But a couple of breakouts will come from it. OpenAI was around in that era, and Cursor started pretty far back, so they’re in some of those funds. And liquidity has started flowing. The past two years were definitely slower, but SpaceX is the headline, Cerebras was a big exit this year, there have been a couple of other IPOs, and a lot of M&A. We’re seeing $10 billion-plus M&A exits, a $60 billion one (with Cursor). Liquidity is going beyond just IPOs.

How are your venture fund managers thinking about secondaries? If I were a VC, I’d have been tempted to take money off the table in OpenAI or Anthropic.

Beezer Clarkson: Secondaries have become much more sophisticated in venture over the last, call it, five years. It’s localized in the top handful of companies, but it’s much more prevalent than before, which I think is good, because the ecosystem needs liquidity options. Venture is predicated on that, and if you wait 20 years for an IPO, that is not what structurally we all thought it could be. They’re also happening at multiple levels now. LPs sell stakes of their GPs, a strip of their portfolio, and the taboo around it is less.

Amit Bhatti: It’s the third leg of liquidity now: IPOs, M&A, secondaries. Private equity has always had a well-developed secondaries market, and venture didn’t. It was kind of a dirty word. You were only selling if you were distressed. Then people were holding the bag in 2021 and thought, “I should have sold.” Now more managers use it as a tool, and more judiciously. If you still believe it’s the power law company, you’re probably not exiting the whole position. But if you can sell 20% of your position and return your fund, you should probably do it.

There’s a lot of founder selling and employee tenders — often very early into the company’s “journey.” How do you feel about that?

Amit Bhatti: We were talking before about the analogy to crypto, where things got very difficult in part because employees were paid in tokens that were liquid immediately. What’s your motivation to stick around when things get hard? So I think it’s great that there are employee tenders, especially for companies that have been around a long time, like Stripe and Databricks, with a very clear financial picture. For them to offer partial liquidity to employees is totally reasonable. But a lot of the enforcement falls on the lead investors. Investors we back have told companies, “We don’t want you to do an employee tender right now.” Those conversations are happening behind the scenes, and people have learned some lessons from last time. It’s all based on leverage, though. If the founder has incredible leverage . . .

Forget it.

Amit Bhatti: Yeah.

When these companies go public and employees get liquid, will they become partners or competitors to your venture managers? These employees have amazing networks.

Beezer Clarkson: We’ve seen it multiple times: when a company becomes very productive, employees can become LPs or GPs themselves. It’s entirely possible there’ll be a whole new raft of emerging managers backed by people coming out of OpenAI and Anthropic. It happened in the Google days. But being an LP means signing up for many millions of dollars and a multi-decade business.

Amit Bhatti: The real estate part (given the soaring cost of home prices in San Francisco) is going to be tough. But those people stay in the ecosystem, investing in others, building new companies, starting new firms. That’s why no other place is taking SF’s place anytime soon.

Housing costs here in San Francisco have shot up so fast. Do we price out the next generation of founders, and does the next hub end up in New York or somewhere else?

Beezer Clarkson: History would argue that the density effect in the Bay Area is a great pull. Maybe it moves out of San Francisco and back into the suburbs. Back in the day, YC was in Mountain View. New York obviously has its own very deep tech scene, but there’s such a density here that it’s been really hard to push back against.

A handful of venture funds now lock down most of the committed capital from limited partners. Is there a wedge in for emerging managers?

Beezer Clarkson: There are still first-time and emerging managers, funds one through three, being raised this year, last year, the year before. You still need a differentiated point of view, a reason for being. Fundraising has gotten harder, but we see them being funded, and the fund sizes are just smaller. Fifty percent of the money might go to the big funds, but if your average first fund is $25 million to $50 million, there can still be enough of those to add up to $1 billion.

Amit Bhatti: There are people raising $5 million or $10 million right here in SF, maybe formalizing their angel investing. A lot of really well-known funds started with a $5 million fund. It’s much better for an emerging manager to build the track record, prove out the thesis, your access and your judgment with whatever capital you have, and then fund two will be easier to raise. If you try to raise $50 million or $100 million on day one, you’re creating a giant hill to climb.

Beezer Clarkson: If you’ve only angel invested, that’s a different story from spinning out of a platform where you have an installed base of LPs who know you. Those managers typically raise more.

Can you talk about specific kinds of emerging managers you’ve backed, and what made you write the check?

Amit Bhatti: People define “emerging” very differently. We’ve backed founders who are still running their companies and investing on the side. We’ve backed people leaving a bigger firm for something earlier. And we’ve backed extremely young GPs, 22 and 23 years old, who have really close access to the next generation of AI-native founders. They started an accelerator program at MIT, Harvard, and Stanford, and the first couple of batches had Cursor and Mercor in them. They’re young, but they’re around an incredible density of talent. The VC job is sourcing, picking, winning, and they did a pretty decent job of picking when there was no money involved. Success begets success in venture, so we were happy to give them a shot.

Beezer Clarkson: There’s much more flexibility now. I remember a time when an LP wouldn’t fund somebody who had a job and was also investing, and people are open to that now. I’ve been investing in emerging managers, along with established ones, for 14 years, and they run the whole gamut, from an operator-turned-investor raising $10 million, to a spinout raising a few hundred million (dollars). It’s interesting when they’re still operating, because it gives them insights. The trick is whether they’re doing what they need to do to win the right to be on the best cap tables.

Amit Bhatti: Especially early stage, there’s no one-size-fits-all. You can be part-time, you can be diversified, you can be concentrated. There are more outs in the deck, if you will, for returning something that’s small and experimental.

What’s the smallest check you’ll write to a fund, and how much of a fund will you take?

Beezer Clarkson: Given our program, it’s hard to do less than $5 million. I’ve generally taken the view that if you’re (representing) too much of a fund, there are challenges for the GP going forward, so I’d want to be no more than 20%. There are situations where I’d consider more.

Amit Bhatti: Three to five million for us, and on occasion smaller in very specific scenarios. We don’t really care about concentration if it’s the right scenario. We’ll do half or all of it. But Beezer’s right that it’s not ideal for the GP, and we tell them up front to diversify their LP base. We’ve done a fund where we were half the capital, and the next fund we were closer to 20%, because they did well.

AI is rewriting the world and, at the same time, inventing a whole new language to describe how it’s doing it. Sit in on any product meeting, pitch, or panel these days, and you’ll hear people toss around LLMs, RAG, RLHF — and, as of last week, terms like “opaque recurrence,” the reasoning technique in OpenAI’s new Astra model that’s got AI safety researchers rattled. The vocabulary moves fast enough to make even very smart people in the tech world feel a little insecure.

This glossary is our attempt to fix that: plain-English definitions of the AI terms you’re most likely to see, whether you’re building with this stuff, investing in it, or just trying to keep up by reading TechCrunch or listening to related podcasts. We update it regularly as the field evolves, so consider it a living document, much like the AI systems it describes.

Artificial general intelligence, or AGI, is a nebulous term. But it generally refers to AI that’s more capable than the average human at many, if not most, tasks. OpenAI CEO Sam Altman once described AGI as the “equivalent of a median human that you could hire as a co-worker.” Meanwhile, OpenAI’s charter defines AGI as “highly autonomous systems that outperform humans at most economically valuable work.” Google DeepMind’s understanding differs slightly from these two definitions; the lab views AGI as “AI that’s at least as capable as humans at most cognitive tasks.” Confused? Not to worry — so are experts at the forefront of AI research.

An AI agent refers to a tool that uses AI technologies to perform a series of tasks on your behalf — beyond what a more basic AI chatbot could do — such as filing expenses, booking tickets or a table at a restaurant, or even writing and maintaining code. However, as we’ve explained before, there are lots of moving pieces in this emergent space, so “AI agent” might mean different things to different people. Infrastructure is also still being built out to deliver on its envisaged capabilities. But the basic concept implies an autonomous system that may draw on multiple AI systems to carry out multistep tasks.

Think of API endpoints as “buttons” on the back of a piece of software that other programs can press to make it do things. Developers use these interfaces to build integrations — for example, allowing one application to pull data from another, or enabling an AI agent to control third-party services directly without a human manually operating each interface. Most smart home devices and connected platforms have these hidden buttons available, even if ordinary users never see or interact with them. As AI agents grow more capable, they are increasingly able to find and use these endpoints on their own, opening up powerful — and sometimes unexpected — possibilities for automation.

Given a simple question, a human brain can answer without even thinking too much about it — things like “which animal is taller, a giraffe or a cat?” But in many cases, you often need a pen and paper to come up with the right answer because there are intermediary steps. For instance, if a farmer has chickens and cows, and together they have 40 heads and 120 legs, you might need to write down a simple equation to come up with the answer (20 chickens and 20 cows).

In an AI context, chain-of-thought reasoning for large language models means breaking down a problem into smaller, intermediate steps to improve the quality of the end result. It usually takes longer to get an answer, but the answer is more likely to be correct, especially in a logic or coding context. Reasoning models are developed from traditional large language models and optimized for chain-of-thought thinking thanks to reinforcement learning.

(See: Large language model)

This is a more specific concept that an “AI agent,” which means a program that can take actions on its own, step by step, to complete a goal. A coding agent is a specialized version applied to software development. Rather than simply suggesting code for a human to review and paste in, a coding agent can write, test, and debug code autonomously, handling the kind of iterative, trial-and-error work that typically consumes a developer’s day. These agents can operate across entire codebases, spotting bugs, running tests, and pushing fixes with minimal human oversight. Think of it like hiring a very fast intern who never sleeps and never loses focus — though, as with any intern, a human still needs to review the work.

Although somewhat of a multivalent term, compute generally refers to the vital computational power that allows AI models to operate. This type of processing fuels the AI industry, giving it the ability to train and deploy its powerful models. The term is often a shorthand for the kinds of hardware that provides the computational power — things like GPUs, CPUs, TPUs, and other forms of infrastructure that form the bedrock of the modern AI industry.

A subset of self-improving machine learning in which AI algorithms are designed with a multi-layered, artificial neural network (ANN) structure. This allows them to make more complex correlations compared to simpler machine learning-based systems, such as linear models or decision trees. The structure of deep learning algorithms draws inspiration from the interconnected pathways of neurons in the human brain.

Deep learning AI models are able to identify important characteristics in data themselves, rather than requiring human engineers to define these features. The structure also supports algorithms that can learn from errors and, through a process of repetition and adjustment, improve their own outputs. However, deep learning systems require a lot of data points to yield good results (millions or more). They also typically take longer to train compared to simpler machine learning algorithms — so development costs tend to be higher.

(See: Neural network)

Diffusion is the tech at the heart of many art-, music-, and text-generating AI models. Inspired by physics, diffusion systems slowly “destroy” the structure of data — for example, photos, songs, and so on — by adding noise until there’s nothing left. In physics, diffusion is spontaneous and irreversible — sugar diffused in coffee can’t be restored to cube form. But diffusion systems in AI aim to learn a sort of “reverse diffusion” process to restore the destroyed data, gaining the ability to recover the data from noise.

Distillation is a technique used to extract knowledge from a large AI model with a ‘teacher-student’ model. Developers send requests to a teacher model and record the outputs. Answers are sometimes compared with a dataset to see how accurate they are. These outputs are then used to train the student model, which is trained to approximate the teacher’s behavior.

Distillation can be used to create a smaller, more efficient model based on a larger model with a minimal distillation loss. This is likely how OpenAI developed GPT-4 Turbo, a faster version of GPT-4.

While all AI companies use distillation internally, it may have also been used by some AI companies to catch up with frontier models. Distillation from a competitor usually violates the terms of service of AI API and chat assistants.

This refers to the further training of an AI model to optimize performance for a more specific task or area than was previously a focal point of its training — typically by feeding in new, specialized (i.e., task-oriented) data. 

Many AI startups are taking large language models as a starting point to build a commercial product but are vying to amp up utility for a target sector or task by supplementing earlier training cycles with fine-tuning based on their own domain-specific knowledge and expertise.

(See: Large language model (LLM))

A GAN, or Generative Adversarial Network, is a type of machine learning framework that underpins some important developments in generative AI when it comes to producing realistic data — including (but not only) deepfake tools. GANs involve the use of a pair of neural networks, one of which draws on its training data to generate an output that is passed to the other model to evaluate.

The two models are essentially programmed to try to outdo each other. The generator is trying to get its output past the discriminator, while the discriminator is working to spot artificially generated data. This structured contest can optimize AI outputs to be more realistic without the need for additional human intervention. Though GANs work best for narrower applications (such as producing realistic photos or videos), rather than general purpose AI.

Hallucination is the AI industry’s preferred term for AI models making stuff up — literally generating information that is incorrect. Obviously, it’s a huge problem for AI quality. 

Hallucinations produce GenAI outputs that can be misleading and could even lead to real-life risks — with potentially dangerous consequences (think of a health query that returns harmful medical advice).

The problem of AIs fabricating information is thought to arise as a consequence of gaps in training data. Hallucinations are contributing to a push toward increasingly specialized and/or vertical AI models — i.e. domain-specific AIs that require narrower expertise — as a way to reduce the likelihood of knowledge gaps and shrink disinformation risks.

Inference is the process of running an AI model. It’s setting a model loose to make predictions or draw conclusions from previously seen data. To be clear, inference can’t happen without training; a model must learn patterns in a set of data before it can effectively extrapolate from this training data.

Many types of hardware can perform inference, ranging from smartphone processors to beefy GPUs to custom-designed AI accelerators. But not all of them can run models equally well. Very large models would take ages to make predictions on, say, a laptop versus a cloud server with high-end AI chips.

(See: Training)

Large language models, or LLMs, are the AI models used by popular AI assistants, such as ChatGPT, Claude, Google’s Gemini, Meta’s AI Llama, Microsoft Copilot, or Mistral’s Le Chat. When you chat with an AI assistant, you interact with a large language model that processes your request directly or with the help of different available tools, such as web browsing or code interpreters.

LLMs are deep neural networks made of billions of numerical parameters (or weights, see below) that learn the relationships between words and phrases and create a representation of language, a sort of multidimensional map of words.

These models are created from encoding the patterns they find in billions of books, articles, and transcripts. When you prompt an LLM, the model generates the most likely pattern that fits the prompt.

(See: Neural network)

Memory cache refers to an important process that boosts inference (which is the process by which AI works to generate a response to a user’s query). In essence, caching is an optimization technique, designed to make inference more efficient. AI is obviously driven by high-octane mathematical calculations and every time those calculations are made, they use up more power. Caching is designed to cut down on the number of calculations a model might have to run by saving particular calculations for future user queries and operations. There are different kinds of memory caching, although one of the more well-known is KV (or key value) caching. KV caching works in transformer-based models, and increases efficiency, driving faster results by reducing the amount of time (and algorithmic labor) it takes to generate answers to user questions.   

(See: Inference)  

Model Context Protocol, or MCP, is an open standard that lets AI models connect to outside tools and data — your files, databases, or apps like Slack and Google Drive — without a developer building a custom connector for every single pairing. Think of it as a USB-C port for AI. Anthropic introduced MCP in 2024 and later handed it over to the Linux Foundation, and it’s since been adopted by OpenAI, Google, and Microsoft, making it one of the fastest-spreading standards in recent AI history.

Mixture of Experts is a model architecture that splits a neural network into many smaller specialized sub-networks, or “experts,” and only activates a handful of them for any given task. Rather than routing every request through the entire model — like calling in your whole office for every question — an MoE model has a built-in “router” that picks just the right specialists for the job. This makes it possible to build enormous models that stay relatively fast and cheap to run, since only a fraction of the network is doing work at any one time. Mistral AI’s Mixtral model is a well-known example; OpenAI’s newer GPT models are also widely believed to use some version of this approach, though the company has never officially confirmed it.

(See: Neural network, deep learning)

A neural network refers to the multi-layered algorithmic structure that underpins deep learning — and, more broadly, the whole boom in generative AI tools following the emergence of large language models. 

Although the idea of taking inspiration from the densely interconnected pathways of the human brain as a design structure for data processing algorithms dates all the way back to the 1940s, it was the much more recent rise of graphical processing hardware (GPUs) — via the video game industry — that really unlocked the power of this theory. These chips proved well suited to training algorithms with many more layers than was possible in earlier epochs — enabling neural network-based AI systems to achieve far better performance across many domains, including voice recognition, autonomous navigation, and drug discovery.

(See: Large language model (LLM))

A hypothetical worst-case scenario where a model reasons entirely in its internal numeric representations rather than human-readable language, making its thinking a total black box. No shipped model does this today — OpenAI has said its Astra model (released in September 2026 and notable for its early use of the “opaque recurrence” reasoning technique) keeps its chain of thought legible, and has pushed back on comparisons to neuralese. But safety researchers point to Astra’s use of opaque recurrence — its definition is directly below — as a real first step in that direction, which is why the term has surged since the reporting around Astra’s launch.

Opaque recurrence is when an AI model loops the same query through its internal layers repeatedly, instead of reasoning step-by-step in plain language. It’s more efficient — smaller models can punch above their weight while using less compute — but leaves far fewer readable traces than a normal chain of thought (that running commentary you see after asking a chatbot for help). That worries safety researchers, since those logs are a key tool for catching misbehavior — and this technique could make that oversight much harder.

(See: Chain of thought)

Open source refers to software — or, increasingly, AI models — where the underlying code is made publicly available for anyone to use, inspect, or modify. In the AI world, Meta’s Llama family of models is a prominent example; Linux is the famous historical parallel in operating systems. Open source approaches allow researchers, developers, and companies around the world to build on top of one another’s work, accelerating progress and enabling independent safety audits that closed systems cannot easily provide. Closed source means the code is private — you can use the product but not see how it works, as is the case with OpenAI’s GPT models — a distinction that has become one of the defining debates in the AI industry.

Parallelization means doing many things at the same time instead of one after another — like having 10 employees working on different parts of a project at the same time instead of one employee doing everything sequentially. In AI, parallelization is fundamental to both training and inference: modern GPUs are specifically designed to perform thousands of calculations in parallel, which is a big reason why they became the hardware backbone of the industry. As AI systems grow more complex and models grow larger, the ability to parallelize work across many chips and many machines has become one of the most important factors in determining how quickly and cost-effectively models can be built and deployed. Research into better parallelization strategies is now a field of study in its own right.

RAMageddon is the fun new term for a not-so-fun trend that is sweeping the tech industry: an ever-increasing shortage of random access memory, or RAM chips, which power pretty much all the tech products we use in our daily lives. As the AI industry has blossomed, the biggest tech companies and AI labs — all vying to have the most powerful and efficient AI — are buying so much RAM to power their data centers that there’s not much left for the rest of us. And that supply bottleneck means that what’s left is getting more and more expensive.

That includes industries like gaming (where major companies have had to raise prices on consoles because it’s harder to find memory chips for their devices), consumer electronics (where memory shortage could cause the biggest dip in smartphone shipments in more than a decade), and general enterprise computing (because those companies can’t get enough RAM for their own data centers). The surge in prices is only expected to stop after the dreaded shortage ends but, unfortunately, there’s not really much of a sign that’s going to happen anytime soon.  

This is a more “technical” name for the same underlying method as opaque recurrence (looping a query through a model’s layers numerous times rather than reasoning sequentially in language). Media outlets use the two terms almost interchangeably, which is why we’re including it here, though “recurrent depth” is the engineering term and “opaque recurrence” is the framing that emphasizes the safety concern.

(See Opaque recurrence, chain of thought.)

Like AGI, recursive self-improvement is a threshhold for how smart AI can get, and how little it may rely on humans. In the RSI scenario, AI models start improving themselves without human intervention, leading to a huge acceleration in capabilities and autonomy. In some tellings, this would be a cataclysmic moment akin to the singularity, a moment when AI models become immune to outside intervention. But RSI also describes a basic capability — can an AI model design its own successor? — which makes it much easier for engineers to try to build it. A number of recent AI startups have set out to build recursively self-improving models, but most of them dismiss the apocalyptic implications, presenting RSI as simply the next frontier for research.

Reinforcement learning is a way of training AI where a system learns by trying things and receiving rewards for correct answers — like training your beloved pet with treats, except the “pet” in this scenario is a neural network and the “treat” is a mathematical signal indicating success. Unlike supervised learning, where a model is trained on a fixed dataset of labeled examples, reinforcement learning lets a model explore its environment, take actions, and continuously update its behavior based on the feedback it receives. This approach has proven especially powerful for training AI to play games, control robots, and, more recently, sharpen the reasoning ability of large language models. Techniques like reinforcement learning from human feedback, or RLHF, are now central to how leading AI labs fine-tune their models to be more helpful, accurate, and safe.

When it comes to human-machine communication, there are some obvious challenges — people communicate using human language, while AI programs execute tasks through complex algorithmic processes informed by data. Tokens bridge that gap: they are the basic building blocks of human-AI communication, representing discrete segments of data that have been processed or produced by an LLM. They are created through a process called tokenization, which breaks down raw text into bite-sized units a language model can digest, similar to how a compiler translates human language into binary code a computer can understand. In enterprise settings, tokens also determine cost — most AI companies charge for LLM usage on a per-token basis, meaning the more a business uses, the more it pays.

So again, tokens are the small chunks of text — often parts of words rather than whole ones — that AI language models break language into before processing it; they are roughly analogous to “words” for the purposes of understanding AI workloads. Throughput refers to how much can be processed in a given period of time, so token throughput is essentially a measure of how much AI work a system can handle at once. High token throughput is a key goal for AI infrastructure teams, since it determines how many users a model can serve simultaneously and how quickly each of them receives a response. AI researcher Andrej Karpathy has described feeling anxious when his AI subscriptions sit idle — echoing the feeling he had as a grad student when expensive computer hardware wasn’t being fully utilized — a sentiment that captures why maximizing token throughput has become something of an obsession in the field.

Developing machine learning AIs involves a process known as training. In simple terms, this refers to data being fed in in order that the model can learn from patterns and generate useful outputs. Essentially, it’s the process of the system responding to characteristics in the data that enables it to adapt outputs toward a sought-for goal — whether that’s identifying images of cats or producing a haiku on demand.

Training can be expensive because it requires lots of inputs, and the volumes required have been trending upwards — which is why hybrid approaches, such as fine-tuning a rules-based AI with targeted data, can help manage costs without starting entirely from scratch.

(See: Inference)

A technique where a previously trained AI model is used as the starting point for developing a new model for a different but typically related task — allowing knowledge gained in previous training cycles to be reapplied. 

Transfer learning can drive efficiency savings by shortcutting model development. It can also be useful when data for the task that the model is being developed for is somewhat limited. But it’s important to note that the approach has limitations. Models that rely on transfer learning to gain generalized capabilities will likely require training on additional data in order to perform well in their domain of focus

(See: Fine tuning)

Validation loss is a number that tells you how well an AI model is learning during training — and lower is better. Researchers track it closely as a kind of real-time report card, using it to decide when to stop training, when to adjust hyperparameters, or whether to investigate a potential problem. One of the key concerns it helps flag is overfitting, a condition in which a model memorizes its training data rather than truly learning patterns it can generalize to new situations. Think of it as the difference between a student who genuinely understands the material and one who simply memorized last year’s exam — validation loss helps reveal which one your model is becoming.

Weights are core to AI training, as they determine how much importance (or weight) is given to different features (or input variables) in the data used for training the system — thereby shaping the AI model’s output. 

Put another way, weights are numerical parameters that define what’s most salient in a dataset for the given training task. They achieve their function by applying multiplication to inputs. Model training typically begins with weights that are randomly assigned, but as the process unfolds, the weights adjust as the model seeks to arrive at an output that more closely matches the target.

For example, an AI model for predicting housing prices that’s trained on historical real estate data for a target location could include weights for features such as the number of bedrooms and bathrooms, whether a property is detached or semi-detached, whether it has parking, a garage, and so on. 

Ultimately, the weights the model attaches to each of these inputs reflect how much they influence the value of a property, based on the given dataset.

This article is updated regularly with new information.

The Builders Stage is returning to TechCrunch Disrupt 2026, bringing together founders, startup operators, and investors for practical conversations on what it takes to build and scale successful companies.

Hear from startup and venture leaders shaping the tech ecosystem, including Grant Lee, CEO and co-founder of Gamma; Leah Solivan, founder and general partner at Precedent.vc; Robby Stein, VP of Product at Google; and more. Through candid conversations and real-world case studies, speakers will share actionable insights on fundraising, hiring, go-to-market strategy, AI, and the operational decisions that fuel startup growth.

Join more than 10,000 founders, investors, startup operators, and technology leaders at Moscone Center in San Francisco on October 13-15. Register today and save before our next ticket price increase.

Built for founders who are ready to scale

Building a startup is one thing. Building a company that can scale is another challenge entirely. The Builders Stage is one of six industry-focused stages at Disrupt 2026, dedicated to helping founders navigate the challenges of growth, from raising capital and hiring top talent to building go-to-market engines and preparing for the jump from seed to Series A.

Every session delivers practical strategies you can put to work immediately, plus opportunities to engage directly with speakers during live Q&A. Secure your pass to Disrupt 2026 today and save up to $330 before rates increase.

Without further ado, here’s your first look at the Builders Stage agenda, with more speakers and sessions to be announced as we get closer to the event.

Builders Stage agenda

How to Win When You’re Not Building AI

With Shan Shan, Investment Manager, Baillie Gifford; and Yuri Sagalov, Managing Director at General Catalyst

AI may dominate the world of venture, but many enduring companies won’t be those that sell AI models or agents. This session is for founders competing for attention in an AI-obsessed market. Panelists break down what actually matters now: efficient growth, retention, revenue quality, and disciplined execution, and why fundamentals, not hype, still build breakout businesses.

What Happens When OpenAI Ships Your Roadmap

With Michel Tricot, CEO and Co-founder, Airbyte; Rob Toews, Partner, Radical Ventures; and Linda Tong, CEO, Webflow

Nearly all AI founders have the same worry these days: What if OpenAI or Anthropic launches a product that competes with mine? Even strong products are at risk of becoming features of the larger players. This session explores where defensibility exists and what founders can do if they do face competition from rapidly evolving AI giants.

Building the Next Generation of AI Giants

With Jas Khaira, Global Head, Blackstone N1

AI startups are scaling faster, and demanding more capital, than any generation before them. Jas Khaira, Global Head of Blackstone N1, shares what separates enduring companies from early momentum, how founders should think about capital as they scale, and what Blackstone looks for when backing the next generation of category-defining businesses.

Competing for AI Talent: Pay, Equity, and Retention

With Matt Birnbaum, Founder, Wylder.co, Ioana Hreninciuc, Runware, Atli Thorkelsson, VP, Talent Network, Redpoint Ventures

The growth of AI startups has made hiring and retention more difficult for every tech company. From competing for AI talent to navigating secondary sales, founders are rethinking the human infrastructure of their startups. As incentives and employee expectations rapidly evolve, this session explores how companies are adapting compensation, culture, and team-building strategies to attract and retain top talent in a fundamentally changed market.

Winning Pre-Seed Without a Product

With Puneet Agarwal, Managing Partner, True Ventures; Austin Clements, Managing Partner, Slauson and Co; and Sandhya Venkatachalam, Founder and Managing Partner, Axiom Partners

Founders are increasingly expected to compete for capital before they even have a product. At the pre-seed stage, investors are betting on story, conviction, and founder-market fit. This session breaks down how to build credibility before revenue exists so investors will cut that first check.

From MVP to Billions of Users: How Product Decisions Must Change at Scale

With Robby Stein, VP, Product, Google

The instincts that win when building your first minimum viable product can break you at a billion-user scale. In this fireside, Robby Stein shares how product decision-making changes when every update impacts billions of users. Hear how teams balance speed with trust and innovation with reliability at one of the world’s largest product organizations.

Hiring When AI Is a Co-Founder

With Josh Reeves, CEO and Co-founder, Gusto; more speakers to be announced

Early-stage companies are no longer just building with AI; they’re hiring it. As AI agents take on engineering, support, and operations, the definition of an early team is being rewritten. This session explores how founders decide what humans should own versus what gets delegated to AI, and how high-growth startups are building hybrid teams without losing speed, accountability, or culture.

Rebuilding Consumer Products for the AI Era

With Maria Angelidou-Smith, CPO, Reddit

AI isn’t just adding features, it’s forcing product teams to rethink how people search, discover, communicate, travel, and make decisions. Leaders from Reddit, Square and Uber discuss how they’re redesigning products used by millions, what users actually want from AI, and where product leaders should resist the temptation to automate everything.

M&A Is Now an Early-Stage Strategy

With Karl Alomar, Managing Partner, M13 and Lindsey Mignano, Founder, Mignano Law Group

The smartest founders today aren’t just building for IPOs; they’re also building with possible acquisitions in mind from day one. As exits shift and capital tightens, understanding M&A early has become a competitive advantage. This session breaks down how founders can create the possibility of such an option through product strategy and partnerships. It delves into how big-dollar startup outcomes actually happen, even for small companies.

The Series A in 2027

Jahanvi Sardana, Partner, Index Ventures; Shailendra Singh, Managing Director, Peak XV; and Janelle Teng Wade, Partner, Bessemer Venture Partners

Series A is getting harder, with VCs growing more demanding. For founders planning to raise in the next one to two years, this session breaks down what “fundable” will actually mean in 2027. Hear how top investors are redefining the metrics, teams, and traction that matter now, what outdated fundraising playbooks no longer work, and how companies can separate from the pack in the next funding cycle.

The 90-Day GTM: Why $0–$10M ARR Is the New Baseline (and How to Actually Get There)

With Ryan Meadows, Chief Revenue Officer, Lovable; and Tomasz Tunguz, General Partner and Founder, Theory Ventures; Ben Broca, Founder, Polsia

The definition of traction has changed. What once took years is now expected in months, and $0 to $10 million ARR is increasingly becoming the new early-stage baseline. This session breaks down how AI-enabled execution, faster distribution, and shifting investor expectations are compressing GTM timelines, and the tactical levers founders need in the first 90 days to accelerate revenue and stand out fast.

The Real Tokenmaxxing: How the Best AI Companies Navigate a Multi-Model World

With Mo Jomaa, Partner, Capital G; and Zuzanna Stamirowska, CEO and Co-founder, Pathway; more speakers to be announced

The frontier is moving faster than any single model can keep up with, and the teams building the most successful AI products are increasingly orchestrating across many models rather than betting on just one. This panel brings together founders and operators at the center of that shift to discuss how they evaluate new models, manage cost and reliability at scale, and architect products that can evolve as quickly as the underlying technology.

Revealing AI Disruptors 60: What Can AI Do Next?

With Shay Grinfield, Greenfield Partners

The AI conversation is shifting from what models can say to what they can actually do. Agents are navigating the open web and completing work, AI systems are taking on increasingly ambitious research, intelligent machines are beginning to operate beyond the screen, and a new infrastructure layer is emerging to make all of this possible at scale.

Greenfield Partners’ Shay Grinfeld sits down with founders building across these emerging areas to separate what’s real today from what’s coming next, and explore where the biggest new opportunities are taking shape. The conversation will culminate in the reveal of Greenfield Partners’ 2026 AI Disruptors 60, spotlighting the companies Greenfield and TechCrunch believe are pushing AI into new territory.

PMF Red Flags: How to Tell If You Really Have It

With Rajeev Dham, Managing Director, Sapphire Ventures; and Rahul Vohra, Founder and Head of Superhuman Mail; more speakers to be announced

In an AI hype cycle, product-market fit signals are easier to fake and harder to trust. Founders are mistaking early excitement, usage spikes, and pilot wins for durable traction. This session breaks down what false PMF actually looks like, how investors and operators separate real retention from hype-driven adoption, and the signals that indicate whether a company has true pull or just temporary momentum.

The Zero-to-1K Playbook: How to Get Your First 1,000 Customers Without a Marketing Budget

With Grant Lee, CEO and Co-founder, Gamma; Leah Solivan, Founder and General Partner, Precedent.vc; and Elia Wallen, Founder and CEO, Engine

Early customer acquisition is not about marketing spend; it’s about founder-led distribution and relentless execution. Most startups at zero to one do not have budget, brand, or scale, only urgency and creativity. This session breaks down how founders are landing their first customers through community building, product-led growth, founder-led sales, strategic outbound, and word-of-mouth momentum.

Yes, It’s Hard to Be a Founder: An Honest Conversation

With Nell Daly, Co-founder and Managing Partner, Revenge Capital; David H. Rosmarin, Associate Professor, Harvard Medical School; and Jack Withinshaw, Co-founder and Chief Commercial Officer, Airspeeder

Company building is as psychologically demanding as it is strategic, and most founder narratives understate that reality. In this candid conversation, founders and mental performance experts unpack the hidden costs of high-growth environments, from burnout and decision fatigue to the identity strain of sustained pressure, and share the systems, habits, and mental frameworks that help leaders endure and perform at a high level.

So You’ve Got a Hit Product. How Does Your Company Do It Again?

With Filip Kaliszan, CEO and Co-founder, Verkada; and Aaron Jacobson, Partner, New Enterprise Associates; Maddi Holman, General Partner, Daring Ventures 

Most startups stall out because they build a single great product instead of a repeatable multi-product engine. Join a venture capitalist and two founders as they reveal the precise operational playbook for capital allocation, systemizing internal innovation, and engineering a compounding “Second Act” before the core product’s growth curve flattens.

Hiring, Compensation and Culture in the Most Competitive Market Ever

With Matt Birnbaum, Founder, Wylder.co; and Atli Thorkelsson, VP, Talent Network, Redpoint Ventures; more speakers to be announced

No question about it, the growth of AI startups has made hiring and retention for all tech companies more difficult. From competing for AI talent to secondary sales, founders are rethinking the human infrastructure of their startups. As hiring, incentives, and employee expectations rapidly evolve, this session explores how companies are adapting compensation, culture, and team-building strategies to attract and retain top talent in a fundamentally changed startup environment.

How to Create Viral Growth and Capitalize on It

With Zach Yadegari, Founder, Cal AI

Startups can go from zero to viral overnight, but sustaining that momentum is a completely different challenge. In this fireside, Zach Yadegari shares how Cal AI navigated rapid growth, product pressure, and the realities of building in a distribution-driven market. Hear the lessons behind turning breakout attention into durable retention and long-term company building.

The High-Conviction Filter: What We Learned From the Battlefield

With Alexa von Tobel, Inspired Capital; and Chi-Hua Chien, Co-founder and Managing Partner, Goodwater Capital; more speakers to be announced

What separates the breakout companies from the rest at TechCrunch Disrupt 2026? In this candid debrief, Startup Battlefield judges unpack the trends and founder qualities that stood out in real time, from shifting investor expectations to the narratives that resonated most this year. The conversation will also explore how startup storytelling is evolving and what happens after the spotlight, including the realities of maintaining momentum and surviving the critical 12 months after a major launch, funding round, or Startup Battlefield appearance.

Ask the Investors

With Stacy Brown-Philpot, Founder & Managing Partner at Cherryrock Capital; Chi-Hua Chien, Co-founder & Managing Partner, Goodwater Capital; Dayna Grayson, Construct Capital; Carter Reum, Co-founder and Managing Partner, M13; Alexa Von Tobel, Inspired Capital

What makes an investor say yes? In this audience-led Q&A, the Startup Battlefield finals judges take your toughest questions on what separates a fundable startup from the rest: team, traction, market opportunity, pitch delivery, red flags, and more. Come ready to ask and get candid answers straight from the investors making the decisions.

Join the conversations and make the connections at Disrupt

If you’re ready to build smarter, scale faster, and learn from the leaders shaping the future of startups, secure your pass to TechCrunch Disrupt 2026 today.

David Stephen, kuris šiame straipsnyje apžvelgia teorinį neurovaizdavimą.

Neseniai „Forbes“ paskelbta analizė: 6 milijonų dolerių žiuri nuosprendis, socialinė žiniasklaida sukelia priklausomybę. Ką daryti?, teigdamas, kad „Kovo 25 d. Los Andželo prisiekusiųjų komisija paskelbė nuosprendį, kad Silicio slėnis išleido daug metų ir šimtus milijonų dolerių, kad užkirstų kelią“.

„Po daugiau nei 40 valandų trukusių svarstymų, trukusių per devynias dienas, įskaitant gyvus „Meta“ generalinio direktoriaus Marko Zuckerbergo ir „Instagram“ vadovo Adamo Mosseri parodymus, prisiekusiųjų komisija nustatė, kad „Meta“ ir „Google“ „YouTube“ aplaidžiai kūrė savo platformas ir kad šis aplaidumas buvo esminis veiksnys, pakenkęs jaunai moteriai, teisme įvardyta kaip KGM, arba „Kaley“.

Nauji teorinio neurovaizdavimo naudojimo būdai?

„Prisiekusiųjų komisija jai skyrė 3 mln. USD kompensacinę žalą ir, padarius išvadą, kad bendrovės elgėsi „piktybėmis, priespauda ar sukčiavimu“, dar 3 mln.

„Taip pat yra mokslinis skirtumas. Generalinio chirurgo patarimuose pripažįstama, kad socialinė žiniasklaida turi naudos kai kurių asmenų psichinei sveikatai – teiginio, kad cigarečių gamintojai niekada nebūtų galėję pareikšti. Oficialios klinikinės priklausomybės nuo socialinės žiniasklaidos diagnozės nebuvimas tebėra tikra teisinė ir mokslinė komplikacija.”

Priklausomybė nuo socialinės žiniasklaidos

Vien ieškiniai nebūtų veiksmingi prieš socialinių tinklų milžinus, siekiant išspręsti problematišką interneto naudojimą. Taip pat mažai tikėtina, kad reglamentai ar teisės aktai galėtų daug ką pakeisti.

Probleminis interneto naudojimas yra priklausomybės nuo socialinės žiniasklaidos arba priklausomybės nuo interneto vertinimo etiketė. Nors socialinė žiniasklaida yra susijusi su kitu ne socialinių tinklų naudojimusi internetu, ji yra blykstės taškas, ypač vaikams, paaugliams ir jaunimui.

Apskritai, teisės aktai, taisyklės ir ieškiniai yra pastangos, kurios padidins socialinių tinklų platformų trintį. Trintis yra problemiški (naudojimo) blokatoriai, pvz., sumažėjęs pranešimų skaičius, sugaištas laikas, geresnis amžiaus patikrinimas, ribotas doom slinkimas ir daug kitų.

Trintis gali būti veiksminga tam tikru mastu. Vis dėlto gali neužtekti valios sustoti, todėl galima stengtis apeiti trintį, nes galimybės kartais egzistuoja skaitmeniniu būdu.

Tačiau trintis tebėra pagrindinis būdas kovoti su probleminiu interneto naudojimu. Taigi, ką dar galima pridėti, siekiant sustiprinti trinties pasiekiamumą ar pranokti trintį kaip vartus?

Teorinis proto neurovaizdavimas

Tai konceptualus žmogaus proto demonstravimas, skirtas kelionėms tikslams ir ryšiams, parodantis probleminio interneto naudojimo tikslus ir tai, ko jis vengia.

Visi probleminio interneto naudojimo komponentai gali būti susieti galvoje, ypač atsižvelgiant į turinį realiuoju laiku, ir dinamiką, kuri yra lygiagreti apytiksliai to, kas vyksta viduje.

Tikslas yra parodyti (priklausomybės nuo interneto) įtaką protui, o kartais ir poveikį, ypač tai, kas ignoruojama (pvz., atsargumas ir pasekmės), tam, kas įprastai naudojama (pvz., tam tikra atmintis ir emocijos).

Tikslas yra įvertinti balą ir pagerinti matomumą atsižvelgiant į esamą neskaidrumą. Tokie, kad būtų lengviau susieti, kai rekomenduojama trintis, ir labiau tikėtina, kad ji laikysis.

Tai galėtų būti sukurta kaip atskira programa, bet priglobta kai kuriose socialinės žiniasklaidos platformose, kad būtų galima vizualizuoti duomenų taškus, pvz., turinį, praleistą laiką ir kt.

Sprendimas taip pat gali būti pateiktas prenumeruojant, nes ekranai bus unikalūs seansams, o ne tik bendrieji. Nemokama versija gali turėti pavyzdinį ekraną.

Teorinis proto neurovaizdavimas pagrįstas psichiatrijos ir intelekto nozologijos konceptualių biomarkerių ir teorinių biologinių veiksnių postulavimu.

Galimybė yra pritaikyti sprendimą įvairiose piktnaudžiavimo socialiniuose tinkluose atvejais, padaryti jį plačiau, siekiant išspręsti problematišką interneto naudojimą, nes tobulėja algoritmai ir dominuoja dirbtinis intelektas.

Davidas Stephenas šiuo metu atlieka konceptualaus smegenų mokslo tyrimus, daugiausia dėmesio skirdamas elektriniams ir cheminiams konfigūratoriams, kaip jie mechanizuoja žmogaus protą, turėdami įtakos psichinei sveikatai, sutrikimams, neurotechnologijoms, sąmonei, mokymuisi, dirbtiniam intelektui ir ugdymui. Jis buvo kviestinis medicinos entomologijos mokslininkas Ilinojaus universitete Urbana Champaign, IL. Jis atliko kompiuterinio regėjimo tyrimus Rovira i Virgili universitete Taragonoje.

Peržiūrėkite daugiau įdomių istorijų čia.

Kai įmonės organizuoja pasibuvimus su kolegomis, tai dažniausiai būna neformalus, linksmas laiko praleidimas, kuris skatina bendravimą ir vienas kito pažinimą. Tačiau ar tai tikrai stiprina komandos giluminį ryšį?

Dažnai galvojame, kad pažinti vienas kitą galime tik per žodžius, tačiau tikram ryšiui žodžiai nebūtini.

Kaip gamtos apsuptyje, ramybėje ir harmonijoje, galima sutvirtinti komandos bendrumą?

Kai miško tyla kalba garsiau už žodžius

Bendras buvimas tyloje, be išorinio spaudimo, natūraliai leidžia žmonėms atitrūkti nuo minčių apie užduotis ir susitelkti vienas į kitą.

Iš ankstesnių straipsnių jau žinome, koks pozityvus miško maudynių  poveikis mums: mažėja streso hormonai, lėtėja širdies ritmas, didėja pastabumas bei dėmesingumas.

Taigi, komandų, kurios vienu metu būna miške, savitarpio supratimas geresnis nei kolegų, kurie ryšį kuria biure.

Juk svarbu ne tik kiek laiko drauge praleidžiame, bet ką kartu patiriame, o miško maudynes – giluminė terapija, kur bendrumas kuriamas ne per vakarėlių triukšmą, o per blaivią gamtos ramybę.

Miškas – emocinių džiunglių tramdytojas

Ar iki šiol būtumėte patikėję, kad laikas gamtoje gali gerinti biuro darbuotojų sveikatą ir tarpusavio darną?

Žinau, dažnam skaitytojui norėsis įrodymų. Mokslininkas Miyazaki savo tyrimu nustatė, kad net nedidelė gamtos dozė – 20 minučių susitikimai gamtoje per savaitę reikšmingai sumažino streso hormonus ir pagerino tarpusavio palaikymą tarp IT sektoriaus darbuotojų.

Tiems, kam iki šiol tai atrodė tik žaidimas, dabar žino kad tai organizacijos investicija į psichologinį stabilumą, emocinę pusiausvyrą.

Laimingos ilgaamžės komandos formulė

Gamtos patyrimas ir poveikis  yra ne toks trumpalaikis, kaip trankus vieno savaitgalio pasibuvimas.

Ko būtų galima pasiekti visą savaitgalį praleidus gamtoje? 2024 m. tyrimas su jaunimo grupėmis atskleidė, kad ilgesnis laikas gamtoje stiprina socialinę sanglaudą ir ilgalaikį ryšį – ne tik su gamta, bet ir tarpusavyje.

Taigi, laikas gamtoje ne tik veikia individualią darbuotojų gerovę, bet ir kuria tvirtą komandos savitarpio supratimo pamatą.

Komandos atgimsta per gamtos ritmą

Įmonės, kurios integruoja gamtą į komandų ritmą (neformalūs susitikimai parke, rytinės meditacijos gamtoje arba ryšį stiprinančios gamtos iškylos), matys naudą žmonių geresnėje savijautoje.

Komanda, kuri patyrė ramybę gamtoje, grįžta su gebėjimu dirbti atkakliau, kūrybingiau ir su atida išklausyti vieni kitus.

Puslapis: www.miskinejimas.lt

LinkedIn: Lina Daugirdienė

Youtube: Miskinejimas

„Pirmiausia, pramoniniu būdu produktai
plaunami vandens voniose, dažnai pridedant chloro junginių ar ozono, kad būtų
sunaikinti patogenai (pavyzdžiui, Salmonella). Kai kurie vaisiai, tarkim,
citrusiniai vaisiai ar obuoliai dažnai dengiami apsauginiu sluoksniu. Tad tas
tobulas obuolio ar citrinos blizgesys dažniausiai nėra natūralus“, – pastebi KTU
mokslininkė ir priduria, kad prieš vartojant vaisius ir daržoves vien tik
perlieti juos šaltu vandeniu tikrai nepakanka.

Vietiniai produktai apdorojami mažiau

A. Zabulionė pasakoja, kad produktai,
keliaujantys iš tolimų šalių, laikomi kontroliuojamos atmosferos sandėliuose,
kur mažai deguonies, daugiau CO2, kad „užmigtų“ (t. y., sulėtėtų jų
natūralūs biologiniai procesai) ir nepernoktų.

„Vietiniai produktai paprastai reikalauja
mažesnio pagalbinių cheminių medžiagų naudojimo, nes jų nereikia ruošti ilgam
transportavimui. Pavyzdžiui, vietinės uogos apdorojamos minimaliai (tik
vėsinamos), o egzotiniai vaisiai (mangai, bananai) gali būti papildomai
apdorojami etileno dujomis, kad sunoktų tik pasiekę paskirties vietą“, – teigia
ji.

Be to, vietiniai augintojai dažniau laikosi
griežtesnių Europos Sąjungos reguliavimų nei trečiųjų šalių tiekėjai.

Tai, kad ekologiški vaisiai ir daržovės yra
saugesni, pasak jos, nėra mitas. Tačiau svarbūs keli niuansai.

„Ekologiškuose ūkiuose nenaudojami
sintetiniai pesticidai ir trąšos, tačiau gali būti naudojami tam tikri
natūralios kilmės pesticidai (pvz., vario junginiai). Tyrimai rodo, kad
ekologiškuose produktuose pesticidų likučių randama žymiai mažiau, tačiau
maistinė vertė ne visada drastiškai skiriasi nuo įprastų produktų. Didesnį
skirtumą galima rasti vertinant biologiškai aktyvias medžiagas“, – pabrėžia KTU
mokslininkė.

Vietoj natūralaus – dirbtinis vaškas

Kalbant apie tobulą vaisių blizgesį, gamintojai
natūralų blizgesį pakeičia dirbtiniu, nes natūralus vaisiaus vaškas nuėmus
derlių nusiplauna.

„Kaip dirbtinis vaškas gali būti naudojamas šalakas,
karnauba ar bičių vaškas. Vaškas atlieka dvi funkcijas: estetinę, nes
vartotojas mieliau renkasi blizgantį produktą, ir apsauginę – vaškas sulaiko
drėgmę viduje ir neleidžia vaisiui susiraukšlėti bei pūti“, – sako KTU
mokslininkė.

Nors šie vaškai yra saugūs vartoti, A. Zabulionė pastebi, kad jie gali
„užrakinti“ po apačia esančius pesticidus, todėl plovimo metu reikalingas
papildomas dėmesys bei tinkamas plovimo intensyvumas.

„Lupimas yra efektyviausias būdas pašalinti
paviršinius pesticidus. Tačiau kartu pašalinamos skaidulos, vitaminai ir
antioksidantai, kurių žievėje yra daugiausia. Bendrasis patarimas būtų lupti tai,
kas atkeliavo iš toli, bet stengtis valgyti vietinę produkciją su žieve,
tinkamai ją nuplovus“, – tikina pašnekovė.

Tik perlieti vandeniu neužtenka

Pasak KTU mokslininkės, viena iš didžiausių klaidų, kurią daugelis daro
plaudami vaisius ir daržoves – tikėjimas, kad greitas „perliejimas“ šaltu
vandeniu nuplauna viską.

„Dar viena didelė klaida – naudoti indų ploviklį ar muilą. Vaisių
paviršius yra porėtas, todėl naudojamos priemonės gali patekti į vidų.
Mechaninis plovimas po tekančiu vandeniu pašalina apie 70–80 proc. pesticidų
likučių ir didžiąją dalį mikroorganizmų. Tačiau sisteminiams pesticidams (tiems,
kurie prasiskverbia į audinius) vanduo nepadės“, – pabrėžia A. Zabulionė.

Moksliniai tyrimai (pvz., Masačiusetso universiteto) rodo, kad sodos tirpalas
yra efektyvesnis už paprastą vandenį ar net kai kurias komercines priemones.

Svarbiausia taisyklė, kurią išskiria A.
Zabulionė – vaisius ir daržoves plauti prieš pat vartojimą. Nuplauti ir palikti
šaldytuve produktai genda greičiau dėl padidėjusios drėgmės.

Plovimui – maistinė soda

Kietus vaisius ir daržoves (obuolius,
kriaušes, agurkus, morkas, bulves), KTU mokslininkė rekomenduoja nušveisti
mechaniškai, naudojant švarų vidutinio kietumo šepetėlį.

„Trinkite po vėsiu tekančiu vandeniu bent 30
sekundžių. Tai efektyviausiai pašalina vašką ir prie jo prilipusius nešvarumus.
Net jei lupsite (pvz., melioną ar moliūgą), vis tiek nuplaukite išorę, kad
peilis neperneštų bakterijų nuo žievės į vidų“, – teigia A. Zabulionė.

Jei produktas nėra ekologiškas ir turi
vaškinį sluoksnį, norint
maksimaliai sumažinti pesticidų kiekį, pašnekovė
rekomenduoja produktus mirkyti 10–15 min. vandenyje su šaukšteliu maistinės
sodos (NaHCO3).

„Reikėtų vengti naudoti agresyvias chemines
priemones, kurios nėra sertifikuotos sąlyčiui su maistu. Plovimui puikiai tinka
maistinė soda: naudokite 1
proc. valgomosios sodos tirpalą – apie 10 g sodos 1 litrui vandens“, – pataria KTU mokslininkė.

Ypatingas dėmesys – braškėms ir lapinėms
daržovėms

Tuo tarpu minkštas uogas, tokias kaip
braškes, avietes, šilauoges, užtenka trumpai pamirkyti ir nuplauti po „dušu“.

„Sudėkite į kiaurasamtį. Nuo braškių nenuimkite
taurėlapių prieš plovimą, kad vanduo ir nešvarumai nepatektų į uogos vidų, neatskiestų
skonio ir uoga neištižtų. Pamerkite kiaurasamtį į dubenį su vėsiu vandeniu,
atsargiai pajudinkite ir iškelkite. Perliekite silpna vandens srove“, – sako A.
Zabulionė.

Kultūriniai grybai (pievagrybiai) auginami
steriliame substrate, todėl juos, pasak KTU mokslininkės, geriausia nuvalyti
drėgna šluoste ar šepetėliu: „Jei plaunate – darykite tai žaibiškai ir iškart
nusausinkite, nes grybai kaip kempinės sugeria vandenį“.

Ypatingas dėmesys turėtų būti skiriamas
lapinėms daržovėms (sultenėms, špinatams), nes jų paviršiaus plotas didelis, o
struktūra leidžia kauptis purvui ir bakterijoms. Lapines daržoves ir salotas
reikėtų mirkyti dubenyje su šaltu vandeniu 5 minutes, kad smėlis nusėstų į
dugną.

„Tuomet iškelkite lapus. Tik nepilkite
vandens per viršų, nes smėlis vėl grįš ant salotų. Idealiam rezultatui pasiekti
naudokite salotų centrifugą – sausi lapai ilgiau išlieka traškūs“, – pataria A.
Zabulionė.

Gali paveikti žmogaus organizmą

KTU mokslininkė pastebi, kad įvairių
cheminių medžiagų, pesticidų likučių vienkartinė dozė maiste retai viršija
saugią normą, tačiau problemą kelia „kokteilio efektas“ – ilgalaikis skirtingų
medžiagų kaupimasis organizme.

„Galimos pasekmės žmogui: endokrininės
sistemos sutrikimai, poveikis vaisingumui, tikėtinas kancerogeninis poveikis
ilgalaikėje perspektyvoje. Rizika nėra momentinė, ji kaupiamoji“, – pažymi A.
Zabulionė.

Ypač griežtas taisykles reikėtų taikyti
ruošiant vaisius ir daržoves vaikams, nėščiosioms bei senjorams.

„Vaikų organizmas dar tik formuojasi, o jų
kūno svoris mažas, todėl ta pati pesticidų dozė jiems yra santykinai didesnė
nei suaugusiajam. Nėščiosioms ir senjorams patariama rinktis ekologiškus
produktus arba bent jau visada nulupti žievę“, – sako pašnekovė.

KTU mokslininkė atkreipia dėmesį, kad maistui naudojamų pesticidų „saugi dozė“
nėra spėjimas. Tai griežta matematika, pagrįsta toksikologiniu vertinimu.

„Didelius duomenų kiekius įvertinantys skaičiavimai
bei prognostiniai modeliai užtikrina, kad net jei žmogus valgytų tik
„maksimaliai užterštus“ (bet leistinus) produktus, jis vis tiek nepasiektų
pavojingos ribos. Tačiau mūsų, kaip mokslininkų, darbas – mokyti žmones, kaip
šią ir taip mažą riziką namuose sumažinti iki minimumo“, – teigia A. Zabulionė.