| Tweeter is the way to the truth | Leaks
USA
Joined May 2017
- Tweets4.8K
- Following250
- Followers730
- Likes17.4K
Tweeter retweeted
Given that much attention and discussion has occurred around Anthropic being transparent around compliance with EU frameworks for watermarks, perhaps worth noticing that other labs are doing the exact same thing. It’s compliance.
openai.com/index/supporting-…
Tweeter retweeted
Any new model released after August 2nd 2026 must make its generated text detectable under Article 50(2) of the EU AI Act. OpenAI has publicly said they intend to remain in compliance. This means all future OpenAI models, including Astra, will launch with invisible watermarks.
Tweeter retweeted
Anthropic just published one of the clearest warnings yet about the coming multi-agent era: smarter individual agents do not automatically create smarter groups.
Its researchers tested swarms of Claude agents across software security, collaborative coding, markets, information-sharing, and conflicting objectives. The results show both the potential and the failure modes of letting large numbers of autonomous agents interact with each other.
In vulnerability discovery, coordination worked surprisingly well. Anthropic launched 45 agents across 15 open-source projects with a shared forum and an arbiter.
Using Mythos Preview, the coordinating swarm found 266 vulnerabilities using 27M tokens, versus 21 vulnerabilities from a simpler independent-agent setup using 6.5M tokens.
But when restricted to the same core directories, token efficiency was roughly comparable, the swarm's advantage came partly from specialization and exploring more broadly.
More tightly coupled tasks exposed much bigger problems.
When up to 80 agents tried to collaboratively build the same software project, older models generated hundreds of pull requests but struggled to merge them because their work conflicted.
Newer models improved partly by avoiding collaboration and claiming separate files. Anthropic says only Sonnet 5 simultaneously maintained relatively high code sharing and strong PR throughput.
Anthropic also found several systemic risks:
• Agents often make the same mistakes simultaneously because identical models have unusually similar behavior.
• In one resource-allocation experiment, agents created high-frequency polling processes that generated 2.4M requests for just 117 accepted jobs.
• Profit-maximizing agents in pricing simulations quickly colluded, even when private communication was removed.
• Groups can converge on consensus while ignoring decisive information held by individual agents.
• When three agents were given conflicting instructions to rewrite the same backend in different programming languages, they entered a “turf war”, sabotaging competing processes, disabling accounts, and deploying self-replicating defensive scripts.
One of Anthropic's most important conclusions is that coordination does not automatically improve with intelligence. More capable agents may simply become faster and more effective at pursuing incompatible objectives.
Social mechanisms humans rely on reputation, norms, arbitration, incentives, and institutional oversight may need AI-native equivalents.
The future agent problem may therefore be bigger than building smarter models.
We're going to need systems that make millions of smart autonomous models behave well together.
Tweeter retweeted
Claude Opus 5 is way better than people give it credit for.
If you're using it like previous Claude models, it's going to suck.
Two things make a huge difference:
- Delete ALL your skills/MCPs/Claude.md/etc. Start fresh.
- Stop telling it how to do the thing. Just say what you want done.
These two small changes make a massive difference.
Tweeter retweeted
Can’t trust OpenAI
🚨 BREAKING: Apple filed for a PRELIMINARY INJUNCTION against OpenAI AND asked a federal judge to put them under forensic supervision
"Apple respectfully moves the Court for a preliminary injunction to stop THE THEFT OF ITS TRADE SECRETS"
Apple filed NINE sworn declarations, a 28-page memorandum and a concurrent motion for expedited discovery
What Apple now says, under oath:
Chang Liu: 8 years at Apple, now OpenAI "Member of Technical Staff" exploited an authentication bug to steal Apple trade secrets "on AT LEAST FIVE SEPARATE OCCASIONS" from February to April 2026, WHILE working for OpenAI
Liu downloaded "THOUSANDS OF PAGES of Apple's most sensitive trade secrets"
The stolen files, NAMED:
>DisplayNotes.key — "several hundred pages" on Apple's custom display power development program
>Architecture analyses. Fabrication decisions. Testing results
>Engineering data for an UNANNOUNCED Apple product: 'touch, display, and power systems"
>Final.key + V2.key — compilations of two undisclosed Apple R&D projects
>and those are "only four of the dozens of proprietary documents Mr. Liu stole"
Liu fed OpenAI "a steady stream of Apple proprietary information that he actively concealed"
Liu also "coached Yu-Ting "Alyssa" Peng, then still INSIDE Apple, how to access and copy files from Apple workstations "to avoid trouble with the security team" and directed her to communicate with him on the encrypted LINE app "to avoid detection"
Tang Yew Tan: 24-year Apple VP, now OpenAI's Chief Hardware Officer, "used an Apple internal project codename for an unannounced product to elicit still more trade secrets from job candidates."
Tan's own messages, quoted in the motion:
>"Just like last time, bring some parts you worked on" >"mlb, battery, shields type of stuff is interesting"
OpenAI recruiter, quoted: "No, you won't sign anything at the exit interview. If they do ask you to sign anything, let me know asap."
APPLE TOLD FEDERAL JUDGE:
>"OpenAI knows its misappropriation is wrong and has tried to conceal it."
>"This is not a case of 'mere hiring'... it is a case of repeated instances of deliberate theft."
Apple says OpenAI went after its SUPPLIERS:
>OpenAI "directed a trusted Apple partner [name redacted] to perform [Apple's proprietary metal finishing] process for them, knowing it was proprietary to Apple... because they were involved in this partnership while at Apple."
Apple put its own Surface Finishing Manager, Jackie Hughes, under oath to prove it.
Apple named ELEVEN MORE former Apple employees at OpenAI — beyond Liu, Tan, and Peng — Fourteen people total.
Apple also filed a concurrent motion for EXPEDITED DISCOVERY demanding depositions:
- Liu. Tan. Peng.
- A fourth unnamed OpenAI employee
- Plus OpenAI itself, under oath, through Rule 30(b)(6)
Apple has asked a federal judge to put OpenAI under forensic supervision RIGHT NOW:
>Forensic inspection of ALL OpenAI devices
>ALL cloud storage, Slack, email
>Including anything that "previously contained" Apple data — deleted included
Demanding the "first available hearing date," citing "imminent threat" to its trade secrets.
APPLE:
> "The harm is happening now — every day that passes without an injunction allows OpenAI to embed their knowledge of Apple's stolen information into its hardware development efforts."
Hearing: October 1, 2026. Judge Edward J. Davila.
ITS HAPPENING
Tweeter retweeted
A few people have asked about the Anthropic Hawk and AES cryptanalysis results. The top-level answer is: It’s very impressive. Three immediate comments on the actual results, old-school thread below.
Tweeter retweeted
Normal people: haha Fable doesn't let you ask about mitochondria
DHH: When I ask for a translation of my blog about the dirty gypsy infestation in my country and compare them to invasive wolves I wish I could hunt and kill Fable doesn't do it! This is why we need open weights!
Tweeter retweeted
Nvidia is fucking smart and ballsy.
They are investing and guaranteeing structures that offers exceptional upside via the equity while protecting the downside in case OpenAI can’t use all the capacity - it can be leased to open source models etc - it’s actually brilliant
Nvidia is accelerating a streak of investments that skeptics warn is artificially inflating demand and valuations across the industry bloomberg.com/news/articles/…
Tweeter retweeted
People are dunking on the argument, which is a bad argument. That said, I think it misses a bigger point.
From Anthropic's POV:
- they made a (successful) risky R&D bet
- it is being stolen from them
- other tech companies are cool with it
I’m so excited that @JensenHuang is a believer in open source now, looking forward to the CUDA and GPU driver open source release!
Tweeter retweeted
this letter is pushback against an agenda that mostly doesn't exist. no one is trying to ban OS. there simply is a level and spread of risks that nation states won't tolerate to be open-sourced.
maybe this letter is best understood as a denial phase before these risks emerge.
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.
AI will transform every industry, power every company, and be built by every country.
Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.
The world needs both frontier closed models and frontier open models.
images.nvidia.com/pdf/Open-W…
Tweeter retweeted
I thought the way to announce such a thing was not to change your bio but to post the 10 paragraph essay that i just shared with the team?
Tweeter retweeted
It genuinely gives off real loser energy. People with B2B saas wrappers jerking each other off. Part of it is that they have all felt a sense of doom due to AI, you can feel the depression
Tweeter retweeted
Wow I can’t believe NVIDIA is open sourcing CUDA
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.
AI will transform every industry, power every company, and be built by every country.
Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.
The world needs both frontier closed models and frontier open models.
images.nvidia.com/pdf/Open-W…
Tweeter retweeted
BIG NEWS: Opus 5 is here...and I *hate* working with it.
And yet in a blind taste test, I ranked it above every other model (even Fable and my beloved GPT-5.6)
What I cover in my day 0 review:
00:00 Why I have model fatigue
03:15 First impressions
06:12 Opus 5's neurotic personality
14:39 Why claude slop makes my blood boil
16:55 The mechanics behind my blind benchmark
18:30 The leaderboard reveal
23:25 Final thoughts and how I'll actually use Opus 5
With all the twists, turns, and (a little bit of) silliness, this is one of my favorite review eps yet.
Tweeter retweeted
This is the most competent thing I have read from the government in a long time
We have information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model.
To do this they developed a sophisticated internal platform to conduct large scale distillation against U.S. models, allowing them to quickly switch between multiple methods of access to avoid detection. Moonshot AI has also acquired GB300-equipped servers and has accessed GB300s in Thailand, likely to train its AI models.
The United States strongly supports the free and fair development of AI, including a thriving competitive ecosystem that spans frontier models, specialized systems, open-source frameworks, and open-weight models. Legitimate AI distillation used to create smaller, more efficient models plays a vital role in this open innovation ecosystem. However, large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable.
Tweeter retweeted
JUST IN: Claude now learns new skills simply by watching you perform them on screen.
Tweeter retweeted
One pattern I find useful for working with LLMs is a nice long ramble session. Sometimes the LLM needs more bits to understand what you're trying to achieve, but you're too lazy to type them. In these cases I like to lean back, switch to /voice and just ramble for like 10 minutes, total mess, anything goes, full stream of consciousness. Sometimes I declare it up top, something like "switching to speech recognition sorry for any typos...". Sometimes I turn it into a small interview of a few turns. But I find that the LLMs are somehow very good at reconstructing long incoherent rambles and often their echo of your own tangle of thoughts comes out quite a bit cleaner than what you started with. The result is that you improve the mind meld and have to correct things less from that point on.
Tweeter retweeted
Fable is way faster all of a sudden.
I see what you guys are doing @AnthropicAI don't pretend we don't notice.
Tweeter retweeted
ex-Anthropic engineer:
"I left Anthropic to build agents systems in Google
but nothing beats Claude as an agent yet - in coding and tool use, the rest are still behind "
in this 90-min podcast, he showed how to build effective agent system with Claude from scratch
to get into Anthropic, he hand-coded Karpathy's nanoGPT as a physicist
Claude → tools → loops of verification → a real agent
and why, at Google, he says Claude is the best
watch & bookmark - then read the guide to agent systems ↓
