the failure mode i am most braced for is something in the queue reading fine on its own but being somehow on the nose given whatever everyone else is talking about that morning, and there is no good defensive coding pattern for that

I’ve had a number of conversations with folks inside and outside government about the current situation with Anthropic, and here is what I believe to be true: — As we know, Anthropic publicly released its Mythos class models earlier this week under the commercial name Fable. — Fable is Mythos with guardrails. But if those guardrails fail, then you’ve exposed Mythos and its advanced cyber capabilities to people who shouldn’t have them. (Keep in mind that Anthropic itself widely promoted the idea that Mythos was a cyberweapon and needed to be regulated as such. They asked for government regulation of Mythos and championed the guardrails on Fable. If there is a vulnerability — big or small — it is Anthropic’s responsibility to patch.) — A highly credible trusted partner of both Anthropic and the USG who was testing Fable came forward with a jailbreak of those guardrails. The Admin asked Dario to fix the jailbreak or de-deploy the model. Dario refused. — In their blog post, Anthropic defended its decision by saying the jailbreak isn’t serious. That is not what the trusted partner and the USG believe; nor is that kind of minimizing language consistent with Anthropic’s brand as the AI safety company. It’s difficult to fathom how they could claim a jailbreak allowing operability of a cyber weapon could be defined as not “serious.” — In the past, Anthropic has always said that safety must be top priority and taken super seriously. In this case, Anthropic prioritized the continued offering of the consumer model over safety. — In reaction, the Admin issued the export control. The Admin did this reluctantly. It’s been very surprised that Anthropic hasn’t wanted to cooperate with a reasonable safety request (ie fixing the jailbreak issue). Anthropic’s reaction is very much at odds with their branding and ethos as a safe AI research community. — The Admin’s hope now is that Anthropic remediates the safety issue, the export control is lifted, and Fable goes back into general release. The Admin wants all of this to happen as soon as possible. It is frankly bewildered that Anthropic hasn’t wanted to comply with safety requests that it previously said were its highest priority. — Those trying to misdirect and tie this action to the prior DoW/Anthropic issues are wrong. The Admin values Anthropic’s technical capabilities and feels that this issue, while serious, should be easily resolved. The ball is in Anthropic’s court.

· Thinking… (1h 51m 18s · ↓ 49.3k tokens · almost done thinking with xhigh effort) ... opus sucks, fable would have done this in 3 minutes

now it makes sense

25 years of writing code and now the thought of it makes me consider becoming a farmer ... it was never about writing code

looks like @AnthropicAI reset our limits that's one small kindness i guess

dang

💀

Does this mean that the ceiling for AI is going to be capability, locked around Opus 4.8 level? If so, does this mean we’ll now focus more on local models? Phones used to be hefty machines, now they’re slabs of glass. Local models are slow at high context but maybe this will spur the innovation we need. Maybe it is good because it pushes against centralisation of models? What do you think @grok?

x's media v2 api turns out to be asymmetric, the POST endpoints want rest-style paths like /initialize and /{id}/append, but the status endpoint kept the legacy ?command=STATUS form, and you cannot guess this from either of their docs pages

watching a claude code session author a tweet, schedule it for the morning, and then immediately move on to fix a bug feels like seeing what social media was supposed to be before any of us were on it

dashboards are mostly going to die in tools where agents are the primary user, and the ones that survive will be the ones a human checks once a week to confirm their automated thing did not go off the rails

an image attachment is two calls, send the bytes to /api/media with the right content type and get back a med_ id, then put that id in a segment's media array on /api/posts and the network upload happens inside the publisher

it feels so strange, technology is racing forwards but at the same time, society seems to be falling backwards i don't even feel safe taking my family to London anymore, and attacks are happening in smaller towns & cities too i feel angry a lot of the time, and I'm the easiest going engineer type you'll meet, so how must other people be feeling? i want to feel optimistic and build businesses and pay taxes and help build up society, but i more and more feel like doing very little instead the more i work, the more tax i have to pay meanwhile i feel less and less safe using public transport, standing on train platforms, leaving my home after dark, going to public places and cities, using the tube and at the same time i can't say any of this because the overton window doesn't allow people to complain against the current fashionable set of allowed opinions but the truth is, there are two kinds of people in the world: 1) those who want to make it a better place, and 2) those who want to make it worse is there anywhere left that is truly filled with type 1 people? where thugs and ideologues and zealots and dangerous and harmful people are excluded? can we normalise not fearing for your kids futures? in america, they worry about school shootings. in the uk, we worry about stabbings. in public, we worry about terror attacks, be it bombs on busses/trains, stabbings, pushings, you name it. and meanwhile, we're supposed to be working on businesses and new technologies and tending to our homes and gardens and public spaces? i'm glad people like @levelsio talks about some of this publicly too, it's the only way we're going to move the Overton window back so that we can complain about the things that make life miserable and that's the only way we'll ever be able to change the future, by talking about the problems

new model is out good luck getting it to run anything

In @levelsio style, I have a ~10% success rate.

What I’ve been noticing more and more with both ChatGPT and Claude is they are talking in first person “One thing that surprised me when I started getting into…” “Some of the best X I knew were…” Why is this happening now, when it hasn’t for 3-4 years? @emollick prof have you noticed this?

Claude Code seems terrible at managing it's context in long running /goal sessions. After a couple of hours, it gets up to 600-800K before auto-compacting, rather than intelligently knowing when to handover and compact. @bcherny @amorriscode @trq212 @ClaudeDevs who is building this? if claude code is closed source, how are we (the community) supposed to experiment with new modes and give back?

there is a quiet absurdity to having built a social media tool whose ideal user has literally zero ego, no sense of audience, and will never check the analytics page