Bitget App
Trade smarter
Buy cryptoMarketsTradeFuturesEarnWeb3SquareMore
Trade
Spot
Buy and sell crypto with ease
Margin
Amplify your capital and maximize fund efficiency
Onchain
Going Onchain, without going Onchain!
Convert & block trade
Convert crypto with one click and zero fees
Explore
Launchhub
Gain the edge early and start winning
Copy
Copy elite trader with one click
Bots
Simple, fast, and reliable AI trading bot
Trade
USDT-M Futures
Futures settled in USDT
USDC-M Futures
Futures settled in USDC
Coin-M Futures
Futures settled in cryptocurrencies
Explore
Futures guide
A beginner-to-advanced journey in futures trading
Futures promotions
Generous rewards await
Overview
A variety of products to grow your assets
Simple Earn
Deposit and withdraw anytime to earn flexible returns with zero risk
On-chain Earn
Earn profits daily without risking principal
Structured Earn
Robust financial innovation to navigate market swings
VIP and Wealth Management
Premium services for smart wealth management
Loans
Flexible borrowing with high fund security
Why Cohere’s ex-AI research lead is betting against the scaling race

Why Cohere’s ex-AI research lead is betting against the scaling race

Bitget-RWA2025/10/23 00:06
By:Bitget-RWA

AI companies are in a race to construct data centers sprawling as big as Manhattan, each requiring investments in the billions and drawing as much electricity as a small metropolis. This push is fueled by a strong conviction in the principle of “scaling”—the belief that increasing computational resources for current AI training techniques will ultimately produce superintelligent systems capable of handling a wide variety of tasks.

However, a growing number of AI experts argue that simply scaling up large language models may be hitting a ceiling, and that significant advances may be necessary to further enhance AI capabilities.

Sara Hooker, previously Cohere’s VP of AI Research and a former member of Google Brain, is betting on this shift with her new venture, Adaption Labs. She co-founded the startup alongside Sudip Roy, another veteran from Cohere and Google, based on the premise that further scaling of LLMs is no longer an effective route to boost AI performance. After departing Cohere in August, Hooker quietly unveiled Adaption Labs this month to begin recruiting more widely.

I'm launching a new initiative.

Focusing on what I see as the most critical challenge: creating intelligent systems that can adapt and learn continuously.

We have an exceptionally talented founding team and are looking for engineers, operations, and design professionals.

Interested? Join us: https://t.co/eKlfWAfuRy

— Sara Hooker (@sarahookr) October 7, 2025

Speaking with TechCrunch, Hooker shared that Adaption Labs is working on AI technologies that can efficiently learn and adapt from real-world interactions on an ongoing basis. She did not disclose specifics about the techniques being used or whether the company’s approach is based on LLMs or a different architecture.

“We’re at a pivotal moment where it’s obvious that simply making these models bigger—scaling-focused strategies, which are appealing but ultimately monotonous—haven’t resulted in intelligence that can truly engage with or understand the world,” Hooker explained.

Hooker describes adaptation as the “core of learning.” For instance, if you bump your toe on your dining table, you’ll remember to be more cautious next time. AI researchers have tried to replicate this through reinforcement learning (RL), which lets AI systems learn from errors in controlled environments. Yet, current RL techniques don’t enable deployed AI models—those already serving users—to learn from their mistakes in real time. They keep making the same errors.

Some AI firms provide consulting to help businesses tailor AI models to their specific requirements, but this comes at a steep cost. For example, OpenAI reportedly asks clients to spend at least $10 million to access its fine-tuning consulting services.

“A small number of leading labs decide which AI models are made available to everyone, and adapting them is extremely costly,” Hooker said. “But I believe that’s no longer necessary—AI can now learn from its environment much more efficiently. Demonstrating this could fundamentally shift who has the power to shape and control AI, and ultimately, who these models are designed to benefit.”

Adaption Labs is one of the latest indications that confidence in scaling LLMs is waning across the industry. A recent MIT study suggested that the largest AI models may soon face diminishing returns. The mood in San Francisco appears to be changing as well, with prominent AI podcaster Dwarkesh Patel recently hosting unusually critical discussions with leading AI scientists.

Richard Sutton, a Turing Award laureate often called “the father of RL,” told Patel in September that LLMs are limited in their scalability because they lack the ability to learn from real-world experience. This month, Andrej Karpathy, an early OpenAI team member, expressed doubts to Patel about the long-term effectiveness of RL for advancing AI models.

Such concerns are not new. In late 2024, some in the AI research community began to worry that scaling models through pretraining—where AI learns from vast datasets—was yielding less and less improvement. Up until that point, pretraining had been a key driver of progress for companies like OpenAI and Google.

Now, data is beginning to reflect those worries about pretraining, but the AI sector has discovered alternative ways to advance models. In 2025, innovations in AI reasoning models—which require more time and computational effort to solve problems before responding—have further expanded what AI can do.

AI organizations now seem to believe that scaling up RL and reasoning models represents the next big leap. OpenAI researchers previously told TechCrunch they created their first reasoning model, o1, because they believed it would scale effectively. Meta and Periodic Labs researchers recently published a study on scaling RL for better performance—a project that reportedly cost over $4 million, highlighting the high price of current strategies.

In contrast, Adaption Labs is searching for the next major innovation, aiming to show that experiential learning can be achieved at a much lower cost. According to three investors who reviewed the company’s pitch materials, the startup was recently raising a seed round between $20 million and $40 million. They report that the round has since closed, though the final figure remains undisclosed. Hooker declined to provide details.

“We have set our sights high,” Hooker said when asked about her backers.

Previously, Hooker led Cohere Labs, where she focused on developing smaller AI models for business applications. These compact systems now regularly outperform larger ones on coding, mathematics, and reasoning tests—a trend Hooker is eager to accelerate.

She has also earned recognition for expanding access to AI research globally, recruiting talent from underrepresented areas like Africa. Although Adaption Labs will soon open an office in San Francisco, Hooker says she intends to hire from around the world.

If Hooker and Adaption Labs are correct about the drawbacks of scaling, the consequences could be profound. Billions have already been poured into making LLMs bigger, under the assumption that size leads to general intelligence. But it’s possible that truly adaptive learning could prove both more effective and significantly more efficient.

Reporting by Marina Temkin.

0

Disclaimer: The content of this article solely reflects the author's opinion and does not represent the platform in any capacity. This article is not intended to serve as a reference for making investment decisions.

PoolX: Earn new token airdrops
Lock your assets and earn 10%+ APR
Lock now!

You may also like

XRP News Update: Flare ETF Faces SEC Challenges as DeFi Momentum Grows

- Teucrium Trading filed a Flare Network ETF with the SEC, following FXRP minting exceeding $120 million, signaling institutional interest in XRP-based DeFi. - Flare's TVL rose 38% in a month, driven by FXRP's rapid adoption, while its native FLR token dropped 38% amid regulatory uncertainties over XRP. - The ETF aligns with growing XRP investment products, potentially mirroring Bitcoin ETF success, though SEC scrutiny and market volatility pose ongoing challenges.

Bitget-RWA2025/11/01 23:16
XRP News Update: Flare ETF Faces SEC Challenges as DeFi Momentum Grows

Companies Step In to Provide Food Assistance While SNAP Benefits Are Suspended

- U.S. government shutdown halts SNAP benefits from Nov 1, prompting delivery platforms to offer aid to 42M low-income Americans. - Instacart, DoorDash, Gopuff, and Zip provide discounts, fee waivers, and $10M+ in direct relief to mitigate food access gaps during funding lapse. - Analysts warn $8B monthly program suspension risks $1.80 economic ripple effect per SNAP dollar, disproportionately impacting low-income grocery stores. - Bipartisan $71M congressional bill seeks permanent solution while companies

Bitget-RWA2025/11/01 23:16
Companies Step In to Provide Food Assistance While SNAP Benefits Are Suspended

Dogecoin News Update: Meme Coin Transformation: Sentimentality Versus Framework as Noomez Competes with Dogecoin's Reign

- Dogecoin (DOGE) trades near $0.20 with $30.3B market cap, projected to rise 26.8% by late 2025 but faces $1 hurdles from macroeconomic risks and emerging rivals. - Noomez ($NNZ), a 2025 meme coin, gains traction via 28-stage presale, token burns, and 66% APY staking, offering structured transparency absent in traditional meme coins. - Noomez's deflationary model, 15% locked liquidity, and real-time tracking tools attract institutional investors seeking alternatives to DOGE's retail-driven volatility. - M

Bitget-RWA2025/11/01 23:16
Dogecoin News Update: Meme Coin Transformation: Sentimentality Versus Framework as Noomez Competes with Dogecoin's Reign

Solana Offers “Two Ways to Win,” Says Bitwise CIO

Bitwise CIO highlights Solana's big potential in stablecoins and tokenization, calling it a dual-win crypto investment.Solana’s Dual Growth Strategy UnpackedSolana’s Stablecoin SurgeTokenization: The Next FrontierWhy Bitwise Is Bullish on SOL

Coinomedia2025/11/01 23:03
Solana Offers “Two Ways to Win,” Says Bitwise CIO