pull down to refresh

From #1462723 yesterday:

Dario Amodei said publicly that Claude is now helping design the next version of Claude itself.

What can possibly go wrong? 🤣

reply

The community note is correct: it was just the Claude Code front-end. But, that is also their main "evidence" that LLM-coded production software is okay: "ship, don't think"

Now, they feel the skill loss. Pun intended. Yes, you can let LLMs code stuff for you, if you instruct them well and you improve the control structure with human review over time. Eventually, the human will not find much issues, because you have created a complete framework. But this requires one to think still. The experienced human backstop isn't a luxury, because humans have curiosity while LLMs have no such trait. That's only simulated, not driven.

So sure, do the LLM thing. But don't be a yoloboi. This is the pain Anthropic feels now. It's an error in execution, by willfully ignoring wisdom gathered through many years of industry experience, because there is too much focus on removing people's jobs, rather than enhancing the result.

"Shipping" isn't a moat on its own - any fool can do that now. The moat now is quality. Ship good shit.

reply

The front end mistake is more evidence that Anthropic is a national security supply chain risk

reply

From a technical perspective, and not having an opinion about the natsec component of your statement, I agree.

reply

Anthropic sued the Department of Defense after their contracts were cancelled and a judge (Rita Lin) agreed with them and placed a temporary injunction on the cancellation.

AI nerds and a female judge presume to understand national security supply chain risks better than Hegseth and Trump

reply

I'd not put my fate in either side of that, if I'm really honest.

reply

Fair enough, the point is national security decisions should be made by elected officials not unelected judges or technocrats

reply
150 sats \ 1 reply \ @optimism 3 Apr

I guess from a natsec pov you're right.

From my low-life pleb position of not being of the ruling class nor voting for them, the problem is even closer to home: Anthropic's software, and similar for all the others, is a supply chain risk, period.

Most corporations in the West are that. They're all leaky af, super easy to coerce if you can press the right buttons, get hacked all the time, fuck up all the time.

Not too long ago I was hired by a security company ran by a retired colonel to advice on some audit results they found at a major telco and what to recommend for a fix. It was... crazy. I told them they were fucking joking. But these people were thorough and professional and they knew what they were talking about. They took shit dead serious, had actual evidence for everything. No storytelling. Provable facts. Not some fuckbois in a suit thinking they know everything because they were able to suck VC dick for half a vision that mutates 360 degrees along the way. This is what I miss in the current tech world. The excellence. It's mostly fuckbois. A casino.