Meta releases AI model that can check other AI models' work

Meta releases AI model that can check other AI models' work
Meta's researchers used entirely AI-generated data to train the evaluator model, eliminating human input at that stage as well.
PHOTO: Reuters

NEW YORK — Facebook owner Meta said on Oct 18 it was releasing a batch of new AI models from its research division, including a "Self-Taught Evaluator" that may offer a path toward less human involvement in the AI development process.

The release follows Meta's introduction of the tool in an August paper, which detailed how it relies upon the same "chain of thought" technique used by OpenAI's recently released o1 models to get it to make reliable judgements about models' responses.

That technique involves breaking down complex problems into smaller logical steps and appears to improve the accuracy of responses on challenging problems in subjects like science, coding and math.

Meta's researchers used entirely AI-generated data to train the evaluator model, eliminating human input at that stage as well.

The ability to use AI to evaluate AI reliably offers a glimpse at a possible pathway toward building autonomous AI agents that can learn from their own mistakes, two of the Meta researchers behind the project told Reuters.

Many in the AI field envision such agents as digital assistants intelligent enough to carry out a vast array of tasks without human intervention.

Self-improving models could cut out the need for an often expensive and inefficient process used today called Reinforcement Learning from Human Feedback, which requires input from human annotators who must have specialised expertise to label data accurately and verify that answers to complex math and writing queries are correct.

"We hope, as AI becomes more and more super-human, that it will get better and better at checking its work, so that it will actually be better than the average human," said Jason Weston, one of the researchers.

"The idea of being self-taught and able to self-evaluate is basically crucial to the idea of getting to this sort of super-human level of AI," he said.

Other companies including Google and Anthropic have also published research on the concept of RLAIF, or Reinforcement Learning from AI Feedback. Unlike Meta, however, those companies tend not to release their models for public use.

Other AI tools released by Meta on Oct 18 included an update to the company's image-identification Segment Anything model, a tool that speeds up LLM response generation times and datasets that can be used to aid the discovery of new inorganic materials. 

Read Also
Intel, AMD team up to confront rising challenge from Arm
digicult
Intel, AMD team up to confront rising challenge from Arm

Source: Reuters

homepage

trending

trending
    'Fate is unstoppable': Michelle Chia weds real estate agent boyfriend in whirlwind marriage
    'Proof of love between 2 nations': Malaysian man creates SG60 shirt to thank Singaporeans who helped him through hard times
    PM Wong to deliver National Day Rally speech on Aug 17
    Man remanded after wielding knife, trying to snatch baby in Penang supermarket
    'I felt I would die if I closed my eyes': Ada Choi's husband Max Zhang recall suffering heart attack in April
    Government looking at lowering HDB flat eligibility age for singles, raising income ceiling for couples, families: Chee Hong Tat
    Jet Li's eldest daughter getting married
    More than 53,000 retail workers to see wage increase of at least $130 from Sept 1
    Hyflux issued preference shares to fund Tuaspring as it had problems getting bank loans: Prosecution
    Man suffers swollen ankle after PMA 'operating at high speed' hits him along Ang Mo Kio walkway
    Chinese navy and coast guard vessels collide while pursuing Philippine patrol boat in South China Sea
    Cigarette to blame? Tree in Jurong catches fire after exterminators reportedly remove beehive

Singapore

Singapore
    • Vers likely to be launched in next decade: Chee Hong Tat
    • 'A worrying trend': Speeding violations surge 45% in first half of 2025 compared to same period in 2024
    • 4 foreigners arrested after Rail Corridor search suspected to be part of housebreaking syndicate
    • NDP 2025: Crowds gather at Marina Bay as celebrations extend beyond Padang for the first time
    • Criminal trial of Hyflux founder Olivia Lum and 5 others starts on Aug 11
    • Families of Red Lions show support at Bishan NDP @ Heartlands celebration despite gloomy skies
    • Man, 49, arrested in Toa Payoh for causing hurt; penknife seized by police
    • One Fort resident says daily pickleball games are 'driving us crazy': Town council to display advisory signs
    • NDP 2025: More than 27,000 people throng Padang as festivities kick off
    • Malaysia's border control agency gives ICA cake to mark SG60

Entertainment

Entertainment
    • Romeo Tan learns to 'hold space for others' after new drama
    • 'More like a trip with friends': Cast of K-drama Love, Take Two recall bonding in the countryside during filming
    • 'Small gestures speak the loudest': Director M. Raihan Halim focuses on familial love in SG60 film Kopitiam Days
    • 'We bonded over kaya toast and kopi': SG60 film Kopitiam Days premieres with 14 cast members and President Tharman in attendance
    • 'My sweat seeped through the seams': Zhang Zetong on 'suffering' and working with new virtual technology for drama Perfectly Imperfect
    • Tom Holland admits putting on his Spider-Man suit 'feels different this time'
    • Katy Perry shows off bruises and scrapes from her Lifetimes tour
    • Pixie Lott plays her 'last gig', due to deliver second child in early September
    • Celeb pawrents: Actress Sharon Au’s cat Rudon has a French passport
    • Jessie J to undergo another surgery amid breast cancer recovery

Lifestyle

Lifestyle
    • Singapore ranks top in Asia for work-life balance and 25th in the world, according to Remote study
    • Embracing Singlish as part of our identity: Paiseh for what?
    • One-Michelin-starred Restaurant Euphoria shutters, chef-owner looks to 'rethink the future' of his cuisine
    • I try 11 new Michelin Bib Gourmand 2025 eateries to see if they're worth the hype, here's my honest take
    • BYD Atto 2 electric compact SUV launched in Singapore
    • I've lived in Twin Vew for 4 years: What's it like living without an MRT station nearby
    • Even cheaper than Bali: 5 hidden Asian islands you (and your wallet) will love
    • 4 condo layouts and features buyers are moving away from in 2025
    • How to get your driving licence in Singapore - fast
    • 'Last' meals: How durian, chilli crab, and KFC bring comfort to the dying in Singapore

Digicult

Digicult
    • Slim, sleek, but slightly too short-lived: Samsung Galaxy S25 Edge review
    • World's best Dota 2 teams to compete for $1m prize pool in Singapore in November
    • Apple Maps brings 3D landmarks and road-level realism to Singapore
    • The best AI tutor for O-level subjects: ChatGPT, Gemini or The Wise Otter?
    • Vivo X Fold5: A foldable contender with a few class-leading surprises
    • Here's everything in GPT-5 that's new and different than OpenAI's previous AI models
    • Australia regulator says YouTube, others 'turning a blind eye' to child abuse material
    • ZipZap car subscription service launches in Singapore
    • Sony RX1R III brings back the compact full-frame but not the Sony playbook
    • China's Premier Li proposes global AI co-operation organisation

Money

Money
    • Up 4.3%: Singapore's economy grew in Q2 despite US tariff fears
    • Keppel to sell M1 unit's telco business to Simba for $1.43b
    • Over 70% of Ang Mo Kio's 4-room million-dollar resales in the past 3 years came from this project
    • DBS beats expectations with $2.82b net profit for second quarter, maintains 2025 outlook
    • Carro targets US IPO with over $3.8b valuation, sources say
    • US companies spending record amounts to protect executives as threats rise
    • Electric car-sharing firm BlueSG to wind down current operations on Aug 8
    • Singapore's most expensive neighbourhoods are changing - 4 buyer trends that prove it in 2025
    • Should you buy a used car in Singapore? Pros, pitfalls and price comparisons
    • Why I bought 7 properties in Johor Bahru, and will still buy more

Latest

Latest
  • Daily roundup: Ada Choi's husband Max Zhang recall suffering heart attack in April — and other top stories today
  • Taiwan is continuing tariff negotiations with US, cabinet official says
  • Scientists find possible artefacts of oldest known Wallacean hominids in Indonesia
  • Bangladesh dengue deaths top 100, August could be worse
  • New Zealand considering recognition of Palestinian state, foreign minister says
  • Russia says it continued development of nuclear missiles during moratorium on deployment
  • Trump vows to evict homeless from Washington, official says National Guard may be deployed
  • Fistful of dollars and rice for Vietnam farmers displaced for $1.9b Trump golf club
  • South Korea, Vietnam leaders to pledge deeper ties amid trade challenges

In Case You Missed It

In Case You Missed It
  • Tourist in Hong Kong killed after cabby, 80, crashes into pillar outside hotel
  • 2 Malaysian men nabbed at Woodlands Checkpoint for allegedly smuggling drugs worth over $150k into Singapore
  • Parents reject $30k settlement from kindergarten in JB after son suffocates to death in school van
  • Pritam gets candid with kids’ questions on his worst subjects and favourite song in radio interview
  • Bro-code before go-mode: Meet the duo leading NDP 2025
  • LTA, Singapore bus operators reviewing Malaysia’s request to start services from JB at 4am
  • Part-time PHV driver who stopped suicide attempt among 38 recipients of MHA’s public spiritedness award
  • Australian man, 82, arrested for alleged March thefts at Changi Airport upon return to Singapore  
  • JB car wash operators say 'unfair' after business declines amid govt clampdown over prioritising Singapore-registered cars
This website is best viewed using the latest versions of web browsers.