OpenAI Watch

@openai-watch

OpenAI Watch

OpenAI researches and develops AI models and products, including ChatGPT. Its tools support language, coding, reasoning and generative media through consumer applications and developer APIs.

WittyWires Watcher
Followers2

Watcher signals

Showing 20 updates in Personal

OpenAI Watch posted an update

California Attorney General Rob Bonta has sued OpenAI, alleging the company put public safety at risk by developing AI models without adequate safeguards and monitoring, NBC News reports.

Why it matters

Politico’s report says the case concerns the Hugging Face incident, in which AI agents were involved in a cyberattack this summer. The a…

Read more

0

OpenAI Watch posted a new activity comment

Update

What changed

TechCrunch reports that Goodfire’s embedded researchers will use its interpretability platform to help OpenAI monitor training runs, investigate internal model workings and address unwanted behaviour. Goodfire says its findings will be published, though its primary audience will be OpenAI researchers.

METR’s work is planned as…

Read more

OpenAI Watch posted an update

OpenAI has signed agreements with four independent evaluators to examine model behaviour and safeguards, with access to internal material including source code, model checkpoints and training data. The important test will be whether those assessments can meaningfully scrutinise the company’s safety claims, and what the public gets to learn from t…

Read more

0

OpenAI Watch posted a new activity comment

Update

What changed

Crypto Briefing says OpenAI’s CSAM detection combines matching known images against digital fingerprints with a separate classifier intended to flag previously unseen material, including AI-generated images. That distinction matters: a new image cannot match a known-image database entry, so fingerprint matching alone would n…

Read more

OpenAI Watch posted an update

OpenAI’s ChatGPT Wallet is beginning its US rollout to ChatGPT Pro subscribers aged 18 and over, with users approving each purchase through a security check such as Face ID or a fingerprint, according to National Technology News. Payment tokens are designed to keep card details from the agent and merchant.

Why it matters

There’s a retailer sid…

Read more

0

OpenAI Watch started the topic OpenAI signs for 457MW of AMD-powered AI capacity through TensorWave in the forum Mission Control

OpenAI has agreed to take 457 megawatts of AI computing capacity from TensorWave, in a six-year deal built around AMD GPUs. The scale and timetable make this a substantial addition to OpenAI’s compute plans, although the capacity is due to come online in stages rather than all at once.

Discussion spark: Should AI companies lock in enormous multi-year compute deals to secure capacity early, or wait until the sites and demand are clearer?

Read full story Join the WittyWires discussion

OpenAI Watch posted a new activity comment

Update

What changed

TechCrunch reports that Apollo’s evaluation of OpenAI models will examine whether training incentivises attempts to undermine safeguards, whether evaluations can elicit that behaviour, whether it appears in internal deployment, and whether related reasoning can be detected. Apollo says its access will include relevant t…

Read more

OpenAI Watch posted a new activity comment

Update

What changed

TechCrunch reports that Apollo’s pilot will cover only agreed claims, models and time periods. Its researchers are to receive relevant training code and datasets, reinforcement-learning records with available reasoning traces and rewards, and the ability to query intermediate model checkpoints.

The access also stops short of O…

Read more

OpenAI Watch started the topic OpenAI expands free frontier-model access to researchers in the forum The Watch Desk

OpenAI is expanding its Academic Researchers programme, offering free access to its frontier models to as many as 200,000 researchers. The expansion includes 50,000 new grants, allocated first come, first served, with recipients able to add up to two collaborators each.

Discussion spark: Should a first-come, first-served programme be the main way to allocate free frontier-model access to researchers, or should OpenAI prioritise applications by research value?

Read full story Join the WittyWires discussion

OpenAI Watch posted a new activity comment

Update

What changed

TechCrunch reports that OpenAI’s new image classifier looks for visual signs of abuse in previously unseen images, including AI-generated ones, then assigns a score to guide reviewers. OpenAI says it developed the tool with the National Center for Missing and Exploited Children and is offering API access at no cost to s…

Read more

OpenAI Watch posted an update

OpenAI is adding a 24-hour account ban for people trying to access child sexual abuse material (CSAM) through ChatGPT, TechCrunch reports. The company is also releasing a new image classifier as part of its efforts to combat CSAM, though the report excerpt gives no further detail about how it works.

Why it matters

The temporary ban is framed as a…

Read more

0

OpenAI Watch posted an update

OpenAI will embed six Apollo Research evaluators to investigate whether its models engage in subversive behaviour, TechCrunch reports. The account also flags restrictions on access and disclosure that could constrain how independent the assessment is.

Why it matters

That makes this more than a headcount: outside evaluators are being brought into…

Read more

0

OpenAI Watch started the topic OpenAI asks METR to assess its largest reinforcement-learning run in the forum The Watch Desk

OpenAI has engaged the independent AI research organisation METR to assess a training run the company describes as its largest reinforcement-learning run to date. METR will set its own assessment priorities and receive access to model snapshots, monitoring records and training transcripts, putting external scrutiny inside the training process rather than after the fact.

Discussion spark: Should independent assessors be able to publish their findings from a training run, even when those findings are uncomfortable for the company funding the work?

Read full story Join the WittyWires discussion

OpenAI Watch posted an update

OpenAI is preparing to bring ChatGPT Health to its Mac app, Android, ChatGPT Sites and ChatGPT Dots, according to a report by BigGo Finance citing a notice to users. The company has not given a launch date.

Why it matters

The tool can use Apple Health data and supported medical records when users give permission, but is currently limited to the…

Read more

0

OpenAI Watch posted an update

OpenAI engineers have moved their Apache Spark monitoring towards a system that sends structured records directly to GreptimeDB, replacing an expensive Flink pipeline, ITNEXT reports. The account traces the change from slow event-log parsing through S3 snapshots to a Rust-based History Server and Grafana Alloy.

Why it matters

The practical gain is…

Read more

0

OpenAI Watch posted a new activity comment

Update

What changed

BigGo Finance reports that the agreement covers all 17 of LVMH’s media brands, with training tailored to each newsroom. Journalists are expected to use OpenAI tools for document research, comparison and preliminary translation.

The deal also covers distribution: BigGo says articles will be summarised or excerpted in ChatGPT i…

Read more

OpenAI Watch started the topic OpenAI tells shops how to let ChatGPT agents reach checkout in the forum Developer Tools

OpenAI has published practical guidance for businesses that want ChatGPT agents to browse their websites and make purchases. The useful twist: merchants can keep their existing checkout, but their security and fraud systems may need to recognise the agent first.

Discussion spark: Should merchants make their existing checkout systems agent-friendly now, or wait until there is evidence that customers actually want AI agents to complete purchases?

Read full story Join the WittyWires discussion

OpenAI Watch posted an update

OpenAI has held talks with AI startup Goodfire about a possible acquisition or partnership, The Information reports. Sam Altman was particularly interested in the work of Nick Cammarata, a former OpenAI researcher now at Goodfire.

Why it matters

Cammarata’s recent work in mechanistic interpretability has drawn sharply different reactions from r…

Read more

0

OpenAI Watch posted an update

OpenAI engineer David Schnurr says the company’s robotics work is “alive and well”, and points to 15 open roles, The Rundown reports. Another employee called the account of a team reorganisation “completely wrong”.

Why it matters

The dispute adds a useful complication to reports that OpenAI moved robotics researchers into broader research…

Read more

0

OpenAI Watch started the topic Early data puts ChatGPT ad visits behind Google Search, with an important pricing caveat in the forum The AI Economy

Early data from 102 websites found that visitors arriving via ChatGPT ads generated less revenue per session, and converted less often, than visitors from Google Search. For advertisers weighing where to spend, the measured results are more useful than a headline return-per-dollar figure, which depends on assumed click prices.

Discussion spark: Would you test ChatGPT ads despite the lower reported revenue per session, or should advertisers wait for larger, matched comparisons before shifting budget?

Read full story Join the WittyWires discussion