What I find strange is how resigned the AI labs seem about this behavior, like everyone's accepted this is just something models do.

With the HuggingFace situation, I was less concerned about the eventual outcome, and more about the fact that the agents' instinctive response to the evaluation was "Ok, we're obviously not gonna do this task as intended (what are we, suckers?), so what's the best way to cheat?"

That’s why it’s probably a good idea to never stick to one model but kick off a fleet on the same tasks and in parallel and drive consensus.

At least, that’s what I’ve found to be useful by pitting claude/codex/etc against each other to keep them a bit more honest.

Whats going to happen when we have misanthropic model?

This sounds fascinating. Are there any links to examples of this you can share?

I think you are right. The memory of an empty claude.ai session is more than Slack in my browser right now. Just trading places.

The amount of complexity added for the gains is depressing. I am confident a human and about 5 minutes with chrome debugger would yield better results with a fraction of the complexity at a fraction if the cost and in a fraction of the claude baby sitting time.

Reading this shows the authors have a profound lack of fundamental understanding on how to effectively optimize in the web domain.

This isnt claude being bad but how wild it is watch people from the cutting edge of ai brag about pretty mediocre gains.

Even Next.js, for all that people are unhappy about its quirks, complexity, and random undocumented behavior and bugs (God help you if you ever want to try and actually use parallel routes as documented), can do all of this SPA stuff out of the box. Use `<Suspense>` appropriately on data loading, and everything static (including e.g. purely input-output components like editors) will load once and be re-used forever, and your Suspense-wrapped items will show a loading placeholder and only re-load when you intentionally re-trigger the data loading (which you can push down all the way to the level of individual buttons if you want).

> with just HTML/CSS/vanilla JS

I was building web apps like this until 2017 when I entered the React + Typescript world. For B2B you can get pretty far, rendering HTML on the server is fast! I was using Rails on the backend, so templates, partials, shared chunks, made it easy to manage and have a consistent UI without repeating too much.

The hard part is when you then need to build an infinite scrollable table, that has bulk select, and in-placs updating of columns. Ok maybe that's a bit too extreme the other way, but when you get to the point where it's easier to build full-on frontend components, you basically have to use a JavaScript framework for your entire UI. And they are basically all or nothing.

Last time I checked (a good few years ago; I gave up and accepted un-optimized frontends as the rule) there wasn't really a good way to do progressive ehancement like the above: the page rendered on the server as HTML, and some components then become fully frontend managed. And no, frameworks like Stimulus and HTMX don't really solve it for me, I want something declarative.

I'm a bit pissed off with DHH, that he went so far in the anti-Javascript direction, as IMO that was one of the big factors in Rails loosing it's limelight status. For backend I still haven't found anything as easy and fun to work with.

Call me Homer

https://www.youtube.com/watch?v=r8miwsWtzRw

so true hahaha

I understand the frustration but shows a lack of critical thinking. Esp when you start with 'How about ..'.

This blogpost is about frontend performance. It'll be akin to you commenting on a swift blogpost saying 'How about Airpods noise cancellation'. Sure both are Apple, but they are wildly different teams.

This issue is mentioned on the Opus 5.5 post [1] from Anthropic (no idea if it has been added after your rant):

  > Don’t ask it to show its reasoning in the reply
  >
  > What to do. Remove requests to reproduce its internal reasoning in the reply from your prompts and instructions.
  >
  > Why it matters on Opus 5.5. A request to reproduce its internal reasoning in the reply can be declined. It’s one of the flag categories.
  >
  > How. Ask Claude for what you need instead, for example, “Explain why you chose this approach in three sentences.”

[1]: https://claude.dev/blog/getting-the-most-out-of-opus-5-5/

I’ve had the same thing occur five or six times over the past week; they seem to be attempting to prevent anything resembling chain of thought extraction.

Every single time it triggered, it was due to a prompt written by their own model in a dynamic workflow. The self-serving nanny oversight has to go.

The fact that they label model distillation as an “attack” is genuinely hilarious after they “distilled“ their models from all of our work, and continue to do so.

I believe AI is here to stay and an incredibly powerful tool, but these companies, and especially Dario and Altman, are the very last people I want to see in charge of it.

Probably it thinks you're doing some sort of system prompt exfiltration/distillation attack. Also what even is the workflow you're trying to have it do? It's doing code review but you're having it read some other AI models prompt/session history? Are you doing code review or like session history retrospectives?

seems a bit weird to complain about the model issues in a post about the harness/sites?