In this week's newsletter, we go through OpenAI's slow, murky march towards being a military contractor. Other details:
- Pentagon officials took meetings at OAI's HQ in 2023
- Palantir approached OpenAI about a military deal in 2024
More here: www.wired.com/story/openai...
New: Sources tell me the Pentagon first experimented with OpenAI's models as early as 2023, but through Microsoft. At the time, OpenAI explicitly banned military use of its AI.
It wasn't clear then, but OpenAI now says that ban never applied to Microsoft.
thank you!
• WIRED for "How to Win a Fight," including "The WIRED Guide to Protecting Yourself From Government Surveillance," by Andy Greenberg and Lily Hay Newman, June 16, and "Print. Fold Share. Download WIRED's 'How to Win a Fight' Zine Here" compiled by Reece Rogers, June 20, at wired.com
So so proud to be a finalist for this year’s national magazine awards, as part of @wired.com’s “How to Win a Fight” package.
You can read the zine I helped compile here:
www.wired.com/story/downlo...
openai in its jestermaxxing era
New: In an interview with WIRED, Greg Brockman tells me his $50M donation to Trump and AI Super PACs was part of an effort to make sure AI benefits all of humanity.
However, he acknowledges AI is not very popular with humanity right now. www.wired.com/story/openai...
OpenAI president on donating $50 million to pro-Trump and AI PACs: "Anything I can do to support this technology benefiting everyone is a thing that I will do...maybe supporting the team is bigger than the people that I happen to be employed with. It's really team humanity.” from @mzeff.bsky.social
OpenAI president Greg Brockman donated millions to super PACs aligned with President Trump -- much to the dismay of many inside the company.
He tells @wired.com that the donations are about being "team humanity."
Great story from @mzeff.bsky.social
SCOOP
Meta has started blocking links to ICE List on Facebook, Instagram and Threads
www.wired.com/story/meta-i...
Google DeepMind staffers are pressing leadership on their plans to safeguard the 3,000-person unit from ICE agents.
"As we have seen, government agency tactics can change and escalate quite rapidly. With offices in many metro areas across the US, are we prepared?"
Scoop from @mzeff.bsky.social
Google's head of security responded internally to clarify the Cambridge incident, though they did not address the broader concerns.
Here's part of the message from a GDM employee, which was sent to the company's roughly 3,000 person AI unit on Monday. More than 20 people reacted with "plus" emojis.
Google's leaders have yet to address it internally, or respond to the ICE incident in Minnesota generally.
New: Google DeepMind staffers are asking company leaders for "plans and policies" to stop ICE from entering Google's campuses, according to internal messages viewed by WIRED
One GDM staffer alleges that a federal agent tried to enter Google's Cambridge campus in the fall.
Silicon Valley is already pouring millions into the US midterm elections, via super PACs looking to elect leaders sympathetic to the AI industry.
From @mzeff.bsky.social
New: OpenAI says it will start testing ads in the United States "in the coming weeks."
The company says ads will NOT influence ChatGPT’s responses, and that it won’t sell user data to advertisers.
more here in @wired.com w/ @zoeschiffer.bsky.social www.wired.com/story/inside...
NEW: More details on the drama at Mira Murati's Thinking Machines Lab.
TML leaders believe its cofounder Barret Zoph engaged in an incident of serious misconduct last year, leading to trust issues between him and Murati.
But now, more researchers are following him and joining OpenAI.
In a memo to staff announcing the news, OpenAI's CEO of Applications Fidji Simo said she doesn't share any of Murati's concerns over Zoph's ethics. Zoph will report to Simo.
Initial scoop about Zoph's firing from @kylierobison.com.
Read more in WIRED: www.wired.com/story/thinki...
NEW: Thinking Machines cofounders Barret Zoph and Luke Metz are back at OpenAI after leaving the ChatGPT-maker less than two years ago.
Zoph told Mira Murati he was considering leaving her startup on Monday. She fired him today, alleging he shared confidential info with competitors, a source said.
big week for new yorkers yelling at people
These documents reveal some of the latest and craftiest ways major AI labs are sucking up data to improve their AI models.
Legal experts say AI labs and contractors seem to be putting themselves at great risk, but the companies may have just decided it's worth it.
This all seems to be part of OpenAI's work to evaluate its AI agents' performance on economically valuable tasks, a key indicator as the company races towards AGI.
In September, the company released GDPval, its first public evaluation towards this goal.
In files we obtained, OpenAI and its data partner, Handshake, tell contractors to remove confidential and personally information in these docs.
OpenAI even directs people towards a public GPT someone made called "Superstar Scrubber" to help erase data. chatgpt.com/g/g-68e84188...
NEW: OpenAI is asking third-party contractors to upload *real work* they've done at current or past jobs to evaluate their AI models.
OpenAI leaves it up to contractors to remove any confidential info.
scoop w/ @willknight.bsky.social and @zoeschiffer.bsky.social
OpenAI VP of research Jerry Tworek is leaving, sources say after OpenAI sided with Chief Scientist Jakub Pachocki in a dispute over its research direction (Maxwell Zeff/Wired)
Main Link | Techmeme Permalink
More here in the latest edition of Model Behavior: www.wired.com/story/openai...
New: OpenAI quietly rolled back ChatGPT's model router for free users last week.
The router—which was costly for the company to serve, but aimed to improve answer quality—was thought to be hurting engagement, one person familiar with the matter told WIRED.
