

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
-
others7 days ago
Here’s What Bitcoin Needs To Do To Confirm Bullish Breakout, According to Trader Who Nailed 2024 BTC Correction – Crypto News
-
Blockchain7 days ago
XRP MVRV Ratio Dips Below The 200-Day MA – Trend Shift Underway? – Crypto News
-
others6 days ago
Bybit Shuts Down Its NFT Marketplace As Crypto Sector Struggles To Recover – Crypto News
-
Technology1 week ago
How to generate Ghibli-style AI portraits using Grok 3 — no ChatGPT subscription needed – Crypto News
-
Technology1 week ago
How to generate Ghibli-style AI portraits using Grok 3 — no ChatGPT subscription needed – Crypto News
-
Business7 days ago
Can Pi Coin Price Hit $1 Soon? – Crypto News
-
others7 days ago
Here’s What Bitcoin Needs To Do To Confirm Bullish Breakout, According to Trader Who Nailed 2024 BTC Correction – Crypto News
-
others7 days ago
Here’s What Bitcoin Needs To Do To Confirm Bullish Breakout, According to Trader Who Nailed 2024 BTC Correction – Crypto News
-
others7 days ago
XRP and Three Other Altcoins Could Witness Another Sell-Off Event, According to Crypto Strategist – Crypto News
-
Cryptocurrency6 days ago
Sony Singapore Now Lets Shoppers Pay in USDC Through Crypto.com – Crypto News
-
Blockchain6 days ago
Dogecoin Breaking These Levels Could Be The Catalyst For Next Bull Run, Analyst Says – Crypto News
-
Business1 week ago
XRP, BTC, ETH Price Prediction As Inflation Data Sparks Downturn in U.S. Stocks – Crypto News
-
Blockchain7 days ago
zkLend hacker claims losing stolen ETH to Tornado Cash phishing site – Crypto News
-
others7 days ago
Gold extends bullish trend amid rising trade tensions; fresh record high and counting – Crypto News
-
Blockchain7 days ago
SpaceX flight bankrolled by crypto investor launches first manned polar orbit – Crypto News
-
Blockchain7 days ago
Will Bitcoin Downtrend Continue? This Metric Suggests Yes – Crypto News
-
Blockchain7 days ago
Binance ends Tether USDT trading in Europe to comply with MiCA rules – Crypto News
-
Blockchain7 days ago
Binance ends Tether USDT trading in Europe to comply with MiCA rules – Crypto News
-
Technology6 days ago
XRP Price Predicted to Reach $10 in April if US Congress Stablecoin Bill Promotes Ripple’s RLUSD – Crypto News
-
Business6 days ago
XRP Price Predicted to Reach $10 in April if US Congress Stablecoin Bill Promotes Ripple’s RLUSD – Crypto News
-
Technology6 days ago
Apple Intelligence debuts on Vision Pro with visionOS 2.4 update: AI-powered features, spatial content and more – Crypto News
-
Cryptocurrency6 days ago
US equities slip after job openings disappointment – Crypto News
-
others6 days ago
Will BNB Price Rally to ATH After VanEck BNB ETF Filing? – Crypto News
-
others6 days ago
Pound Sterling consolidates against US Dollar ahead of Trump’s tariffs announcement – Crypto News
-
Cryptocurrency6 days ago
FLOKI price poised for 20% rally, Here’s why – Crypto News
-
others6 days ago
PENDLE Price Jumps 8% Today Amid Huge Whale Accumulation – Crypto News
-
others6 days ago
Fundstrat’s Tom Lee Calls for Imminent Stock Market Reversal, Says US Has the ‘Right Pieces’ for a Bottom – Crypto News
-
Technology6 days ago
European regulators warn of financial risks from US crypto integration – Crypto News
-
Business1 week ago
Builder.ai Announces Third-Party Audit After Allegations – Crypto News
-
Technology7 days ago
Over 60 pc broadband, fiber, DSL users surveyed flag problems with connection: LocalCircles poll – Crypto News
-
Cryptocurrency7 days ago
Tether Boosts Bitcoin Holdings By 8,888 BTC In Q1 2025 – Crypto News
-
others7 days ago
Austria Unemployment fell from previous 347.4K to 316.3K in March – Crypto News
-
Blockchain7 days ago
Hayes Predicts $250,000 Bitcoin As Fed Caves To QE Pressure – Crypto News
-
Cryptocurrency7 days ago
XRP, DOGE Shoot up as BTC Price Reclaims $84K Level (Market Watch) – Crypto News
-
Technology7 days ago
Best washing machines under ₹10000 in April 2025 to boost your laundry routine without overspending – Crypto News
-
Blockchain7 days ago
Analyst Calls Dogecoin Chart A ‘Beauty’ As Key Indicators Align – Crypto News
-
Blockchain6 days ago
Bitcoin Price Bounces Back—Can It Finally Break Resistance? – Crypto News
-
Cryptocurrency6 days ago
US equities slip after job openings disappointment – Crypto News
-
Blockchain6 days ago
Several Altcoins Crash Up To 50% On Binance, What’s Going On? – Crypto News
-
Business6 days ago
What to Expect From XRP Price as Trump’s ‘Liberation Day’ Tariffs Go Into Effect Today – Crypto News
-
Metaverse1 week ago
The Tools of Tomorrow: What Lies Ahead with the AI Revolution – Crypto News
-
Business1 week ago
MicroStrategy Acquires 22,048 Bitcoin For $1.92 Billion – Crypto News
-
Business7 days ago
ChatGPT To Launch Next Big Model, New Studio Ghibli Ahead? – Crypto News
-
others7 days ago
Will the RBA hint at further interest rate hikes at its policy meeting? – Crypto News
-
Technology7 days ago
Bumper discounts in Amazon Gaming Fest! Up to 70% off on gaming laptops, monitors, vlog cameras and more – Crypto News
-
Technology7 days ago
Whale Offloads ETH at Loss, But Experts Predict Ethereum Price Rally Amid April Macroeconomic Events – Crypto News
-
Technology7 days ago
SBI down: Mobile banking, ATMs and more affected; here’s what the bank said… – Crypto News
-
Cryptocurrency7 days ago
Crypto Game ‘Blade of God X’ Accused of Mismanagement by Former Exec – Crypto News
-
others7 days ago
GBP/USD could stretch lower if 1.2900 support fails – Crypto News
-
others7 days ago
Goldman Sachs Raises Chance of Recession in a Year to 35%, Says Trump Tariffs To Cause Inflation Spike: Report – Crypto News