Skip to content

Topic

AI Safety Debate

The running public and industry argument over what risks advanced AI systems pose and how governments and labs should respond.

Current stories

investConfirmed2 publishers

OpenAI plans $856bn of compute against $840bn of revenue through 2030

The Financial Times reported a July company presentation that keeps OpenAI free-cash-flow negative every year to 2030, and on its own pace the money raised in March covers only the first three years of that.

Perspective Coverage

4 publishers
Builder
Builder 9%
Operator
Operator 17%
Investor
Investor 74%

Reality

Evidence55
Adoption55
Hype gap+15
Incentives78
Confidence60
investOne report1 publisher

Salesforce runs Agentforce a generation behind the models its keynote guests debated

Nvidia's Jensen Huang told frontier labs to run as fast as they can. On the Dreamforce floor, the Salesforce customers and partners CNBC quoted said last year's models already cover the work they sell, and one of them would welcome a pause.

Publishers:cnbc.com

Reality

Evidence58
Adoption45
Hype gap+18
Incentives72
Confidence55
investOne report1 publisher

Goldman fell nearly 8.5% in the week the Nasdaq gained 0.7%

The Fed's first hike in three years took its benchmark to 3.75%-4% and the Dow down 1.7% for a third straight losing week. Chipmakers and data center suppliers sold off Monday on AI safety headlines and got most of it back by Friday.

Publishers:cnbc.com

Reality

Evidence50
Adoption
Not scored
Hype gap+22
Incentives78
Confidence45
investOne report1 publisher

OpenAI's $1.2 trillion round asks $348 billion more than its March price

The Wall Street Journal reports OpenAI raising at $1.2 trillion, up from $852 billion in March, and the memory shortage that money is meant to buy through is drawing SK Hynix toward Intel's planned Ohio fab.

Reality

Evidence34
Adoption28
Hype gap+32
Incentives72
Confidence42
buildOne report1 publisher

Debate's oversight guarantee rests on a decomposition the model has to invent itself

Arcadia's researchers pitted math and coding against a text-based Geoguessr proxy, and report no uplift from debate training on the two fuzziest tasks they ran, using Best-of-N as the stand-in.

Publishers:lesswrong.com

Reality

Evidence34
Adoption12
Hype gap+22
Incentives35
Confidence50
leadershipOne report1 publisher

A DeepSeek engineer says AI will probably match or surpass the kernels he writes by hand

Liu Shengyu, who helped write a component of DeepSeek's latest model, wrote that AI-written operators will probably be as good as his own or better. He expects to keep his job and to lose the hands-on part of it.

Reality

Evidence48
Adoption20
Hype gap+24
Incentives66
Confidence57
leadershipOne report1 publisher

OpenAI Raised the AI Safety Alarm - Now It and Anthropic Are Selling Cyber-Defense

Newcomer reports founders swapping security consultants for frontier models while OpenAI stages a cyber session on Astra launch day. Neither lab has put a revenue number on the category.

Publishers:newcomer.co

Reality

Evidence44
Adoption38
Hype gap+26
Incentives79
Confidence48