If you feel for 100,000 Gazans then you should also feel for the 375,000 Yemenis (85,000 of them children who starved to death due to the Saudi blockade).
It’s an astounding amount of hypocrisy from the labs to have gotten their initial training data in the way that they did, burning it as a source for others with how much bot protection now exists, and then trying to lift the ladder up after them by preventing distillation
Reading this makes me realize that buying AI inference means placing a lot of trust in the company selling it.
Who at Anthropic can read my conversations, under what conditions, and who oversees those decisions? What determines whether private customer activity becomes material for a public report?
Preventing abuse matters, but I'm uncomfortable with the same company selling the service, judging (and lecturing) its customers, and deciding what to disclose about them.
As a paying customer, I get a bot when I need support. Meanwhile, the company publishes detailed reports about investigating how customers use its product. It feels like a restaurant where I can't get any waiter's attention, but management is very interested in observing how I eat.
We found another cyberattack by internal OpenAI agents, this time targetting @rubygems.
They:
1) gained arbitrary remote code execution on rubydoc.
2) developed a novel exploit to steal user API keys (but we do not know if they succeeded).
They used package names including hack.rb, evil.rb, inject.rb, and exploit.rb.
We thank @j0wimo for initially discovering that agents had posted to RubyGems.
Fable and Mythos 5.1 are the EXACT same weights, they look at the internal activations then escalate to a bigger classifier and ultimately fallback to Opus 4.8 if the request is categorized as dangerous
Fable is not distilled version of a larger Mythos model
@jamonholmgren Overall across the SDLC I see these models as power tools and yes there will be expertise and a certain etiquette to how they're used but every now and again you need to go in with a scalpel, more so with RN/native/mobile than web.
@jamonholmgren Post Opus 4.6 my overall productivity uplift across all tasks is around +15% - 35%. Newer models typically need course correction for the initial 20% then I pretty much wait for the output. Other than that the night shift style usage is bulletproof for me, what about you?
@SaadInCyber Engines and tooling help, but you can spot an Unreal or Unity game from a mile away. Still shipping even a decent experience is very hard and is the most expensive software in the world. Steam probably has millions of dead inventory that doesn't sell.
I used to play and coach a semi pro football team, once we had a prospect come in whose family member played for the Saudi NT, and the specimen I saw that day really put into perspective that there are unavailable levels to this game, @PlanetScale is that guy.
planetscale doesn't exist
i waited to collect full 4 months running both dev and prod entirely on it before saying anything.
guess what. no feedback, no issues, no downtime, no nothing. running your db on planetscale is like not having a db at all.
first time in over a decade i've had a service i never need to look at. it just works.
pristine tier service. and i'm probably one of the least profitable clients they have.
@planetscale@samlambert and company are actually goated.
Idk if it's because pakistan has the second largest shia population or because our sunni scholars have generally been more sensible but i've never come across a pakistani named Yazid or Muawiya. I've met Arabs with both names though
After reading more history, I've come to think that land reform is one of the biggest evolutionary transitions for a civilization, and a great filter. If you can't do land reform, you're doomed to mediocrity or catastrophe. If you can, almost anything else will be forgiven