

Metaverse
Researchers are figuring out how large language models work – Crypto News
LLMs are built using a technique called deep learning, in which a network of billions of neurons, simulated in software and modelled on the structure of the human brain, is exposed to trillions of examples of something to discover inherent patterns. Trained on text strings, LLMs can hold conversations, generate text in a variety of styles, write software code, translate between languages and more besides.
Models are essentially grown, rather than designed, says Josh Batson, a researcher at Anthropic, an AI startup. Because LLMs are not explicitly programmed, nobody is entirely sure why they have such extraordinary abilities. Nor do they know why LLMs sometimes misbehave, or give wrong or made-up answers, known as “hallucinations”. LLMs really are black boxes. This is worrying, given that they and other deep-learning systems are starting to be used for all kinds of things, from offering customer support to preparing document summaries to writing software code.
It would be helpful to be able to poke around inside an LLM to see what is going on, just as it is possible, given the right tools, to do with a car engine or a microprocessor. Being able to understand a model’s inner workings in bottom-up, forensic detail is called “mechanistic interpretability”. But it is a daunting task for networks with billions of internal neurons. That has not stopped people trying, including Dr Batson and his colleagues. In a paper published in May, they explained how they have gained new insight into the workings of one of Anthropic’s LLMs.
One might think individual neurons inside an LLM would correspond to specific words. Unfortunately, things are not that simple. Instead, individual words or concepts are associated with the activation of complex patterns of neurons, and individual neurons may be activated by many different words or concepts. This problem was pointed out in earlier work by researchers at Anthropic, published in 2022. They proposed—and subsequently tried—various workarounds, achieving good results on very small language models in 2023 with a so-called “sparse autoencoder”. In their latest results they have scaled up this approach to work with Claude 3 Sonnet, a full-sized LLM.
A sparse autoencoder is, essentially, a second, smaller neural network that is trained on the activity of an LLM, looking for distinct patterns in activity when “sparse” (ie, very small) groups of its neurons fire together. Once many such patterns, known as features, have been identified, the researchers can determine which words trigger which features. The Anthropic team found individual features that corresponded to specific cities, people, animals and chemical elements, as well as higher-level concepts such as transport infrastructure, famous female tennis players, or the notion of secrecy. They performed this exercise three times, identifying 1m, 4m and, on the last go, 34m features within the Sonnet LLM.
The result is a sort of mind-map of the LLM, showing a small fraction of the concepts it has learned about from its training data. Places in the San Francisco Bay Area that are close geographically are also “close” to each other in the concept space, as are related concepts, such as diseases or emotions. “This is exciting because we have a partial conceptual map, a hazy one, of what’s happening,” says Dr Batson. “And that’s the starting point—we can enrich that map and branch out from there.”
Focus the mind
As well as seeing parts of the LLM light up, as it were, in response to specific concepts, it is also possible to change its behaviour by manipulating individual features. Anthropic tested this idea by “spiking” (ie, turning up) a feature associated with the Golden Gate Bridge. The result was a version of Claude that was obsessed with the bridge, and mentioned it at any opportunity. When asked how to spend $10, for example, it suggested paying the toll and driving over the bridge; when asked to write a love story, it made up one about a lovelorn car that could not wait to cross it.
That may sound silly, but the same principle could be used to discourage the model from talking about particular topics, such as bioweapons production. “AI safety is a major goal here,” says Dr Batson. It can also be applied to behaviours. By tuning specific features, models could be made more or less sycophantic, empathetic or deceptive. Might a feature emerge that corresponds to the tendency to hallucinate? “We didn’t find a smoking gun,” says Dr Batson. Whether hallucinations have an identifiable mechanism or signature is, he says, a “million-dollar question”. And it is one addressed, by another group of researchers, in a new paper in Nature.
Sebastian Farquhar and colleagues at the University of Oxford used a measure called “semantic entropy” to assess whether a statement from an LLM is likely to be a hallucination or not. Their technique is quite straightforward: essentially, an LLM is given the same prompt several times, and its answers are then clustered by “semantic similarity” (ie, according to their meaning). The researchers’ hunch was that the “entropy” of these answers—in other words, the degree of inconsistency—corresponds to the LLM’s uncertainty, and thus the likelihood of hallucination. If all its answers are essentially variations on a theme, they are probably not hallucinations (though they may still be incorrect).
In one example, the Oxford group asked an LLM which country is associated with fado music, and it consistently replied that fado is the national music of Portugal—which is correct, and not a hallucination. But when asked about the function of a protein called StarD10, the model gave several wildly different answers, which suggests hallucination. (The researchers prefer the term “confabulation”, a subset of hallucinations they define as “arbitrary and incorrect generations”.) Overall, this approach was able to distinguish between accurate statements and hallucinations 79% of the time; ten percentage points better than previous methods. This work is complementary, in many ways, to Anthropic’s.
Others have also been lifting the lid on LLMs: the “superalignment” team at OpenAI, maker of GPT-4 and ChatGPT, released its own paper on sparse autoencoders in June, though the team has now been dissolved after several researchers left the firm. But the OpenAI paper contained some innovative ideas, says Dr Batson. “We are really happy to see groups all over, working to understand models better,” he says. “We want everybody doing it.”
© 2024, The Economist Newspaper Limited. All rights reserved. From The Economist, published under licence. The original content can be found on www.economist.com
-
Blockchain1 week ago
Robinhood Dealing With Fallout of Tokenized Equities Offering – Crypto News
-
Cryptocurrency1 week ago
Whale Sells $407K TRUMP, Loses $1.37M in Exit – Crypto News
-
Cryptocurrency1 week ago
Satoshi-Era Bitcoin Whale Moves Another $2.42 Billion, What’s Happening? – Crypto News
-
Blockchain7 days ago
Ripple and Ctrl Alt Team to Support Real Estate Tokenization – Crypto News
-
Technology7 days ago
Fed Rate Cut Odds Surge As Powell’s Future Hangs In The Balance – Crypto News
-
Cryptocurrency1 week ago
Cardano’s $1.22 target: Why traders should be aware of THIS ADA setup – Crypto News
-
Cryptocurrency1 week ago
Bitcoin Breaches $120K, Institutional FOMO Takes and House Debate Propel Gains – Crypto News
-
Technology7 days ago
Fed Rate Cut Odds Surge As Powell’s Future Hangs In The Balance – Crypto News
-
Cryptocurrency1 week ago
Why Is Bitcoin Up Today? – Crypto News
-
Blockchain1 week ago
UK Banks Should not Issue Stablecoins – Crypto News
-
Cryptocurrency1 week ago
Donald Trump Jr. backs social media startup aiming to become a crypto powerhouse – Crypto News
-
Blockchain1 week ago
The Bitcoin Liquidity Supercycle Has Just Begun: Hedge Fund CEO – Crypto News
-
Cryptocurrency1 week ago
Strategy Resumes Bitcoin Buys, Boosting Holdings to Over $72 Billion in BTC – Crypto News
-
Business1 week ago
Pepe Coin Rich List June 2025: Who’s Holding Highest PEPE as it Nears Half a Million Holders? – Crypto News
-
Cryptocurrency1 week ago
It’s a Statement, Says Bitfinex Alpha – Crypto News
-
others7 days ago
EUR/USD recovers with trade talks and Fed independence in focus – Crypto News
-
Cryptocurrency6 days ago
Bitcoin trades near $119K after new all-time high; Coinbase rebrands wallet to ‘Base App’ – Crypto News
-
Business6 days ago
XLM Is More Bullish Than ETH, SOL, And XRP, Peter Brandt Declares – Crypto News
-
Technology1 week ago
Google, Anthropic, OpenAI and xAI join US defence to tackle national security with AI – Crypto News
-
Cryptocurrency1 week ago
Stellar [XLM] bulls exhausted after rally – Is a pullback nearby? – Crypto News
-
Cryptocurrency1 week ago
Fed’s Hammack Raises Inflation Concerns Amid Push For Interest Rate Cut – Crypto News
-
Cryptocurrency7 days ago
1inch price forecast: 1INCH hits 7-month high after double digit gains – Crypto News
-
Cryptocurrency7 days ago
1inch price forecast: 1INCH hits 7-month high after double digit gains – Crypto News
-
others7 days ago
Top Crypto Exchange by Trading Volume Binance Announces Airdrop for New Ethereum (ETH) Ecosystem Altcoin – Crypto News
-
others6 days ago
VanEck Details Key Drivers Boosting Bitcoin Price, Including Corporate Treasury Demand, ETF Flows and More – Crypto News
-
Business6 days ago
XRP Lawsuit Update: Ripple Paid $125M in Cash, Settlement Hinges on Appeal – Crypto News
-
Cryptocurrency5 days ago
Anarchy, crime and stablecoins – Blockworks – Crypto News
-
Business1 week ago
CME XRP Futures Hit $1.6B In Total Trading Volume Since Launch – Crypto News
-
Technology1 week ago
V Guard INSIGHT-G BLDC fan review: Cool performer with a premium look – Crypto News
-
Technology1 week ago
XLM Price Forecast: Why Stellar Lumens May Crash After 80% Rally in Last 7 Days – Crypto News
-
Metaverse1 week ago
Why voice is emerging as India’s next frontier for AI interaction – Crypto News
-
Metaverse1 week ago
Nvidia’s Jensen Huang says AI ‘fundamental like electricity’, praises Chinese models as ‘catalyst for global progress’ – Crypto News
-
Business7 days ago
Ethereum Price Prediction- Bulls Target $3,700 As ETH Treasury Accumulation Soars – Crypto News
-
others6 days ago
GBP/USD rallies on US PPI dip and Trump’s potential Powell removal – Crypto News
-
others6 days ago
GBP/USD rallies on US PPI dip and Trump’s potential Powell removal – Crypto News
-
De-fi1 week ago
Robinhood Opens Ether and Solana Staking to US Users – Crypto News
-
Cryptocurrency1 week ago
Top 3 altcoins under $1 worth watching: Sei, Ethena, Arbitrum – Crypto News
-
De-fi1 week ago
Ripple’s RLUSD Market Cap Passes $515M, Flips TrueUSD – Crypto News
-
others1 week ago
Bitcoin Critic Vanguard Becomes Strategy’s (MSTR) Largest Shareholder – Crypto News
-
De-fi1 week ago
DeFi TVL Surges Past $126B, Up Over 45% Since April – Crypto News
-
Cryptocurrency1 week ago
Nothing Burger or Crypto Catalyst? – Crypto News
-
Cryptocurrency1 week ago
Pi Network Activity Soars as Mystery Wallet Acquires Millions in PI Coins – Crypto News
-
De-fi6 days ago
U.S. Marshals Peg Federal Bitcoin Holdings at 28,988 Tokens Worth $3.4 B – Crypto News
-
Cryptocurrency6 days ago
Russia’s $85 Billion Sberbank to Launch Crypto Custody Services – Crypto News
-
Blockchain6 days ago
Nasdaq Exchange Files SEC Form to List Staking Ethereum ETF – Crypto News
-
De-fi5 days ago
U.S. House Passes Clarity, GENIUS, and Anti-CBDC Acts With Historic Bipartisan Support for Crypto – Crypto News
-
Technology5 days ago
Malicious code found in fake coding extensions used to steal crypto – Crypto News
-
Cryptocurrency5 days ago
XRP Price Spikes to Record Highs As Momentum Signals Extended Gains – Crypto News
-
Cryptocurrency4 days ago
US House passes three key crypto bills; market reaction muted as Bitcoin dips – Crypto News
-
Technology4 days ago
“Decentralized Ponzi Scheme”- Gold Bug Peter Schiff Slams Landmark Crypto Bills – Crypto News