stldev

The METR report continues to hold up. I would add "No Silver Bullet" to the reading list.

Careful who you share this information with- better to roll with the kool-aid drinkers when they're holding the cards.

I think it's natural to lose enthusiasm over time when a joy becomes a job.

"I don't care for coding new stuff. Everything I may need either already exists or is too complex to do on my own (and no, I won't vibe-code it, what's the fun in that?)"

I'm not sure if you mean "code gen without a plan/expertise" or just code gen. If you found joy because you enjoyed building things, now be the best time to explore and prototype something you've always dreamt of.

If you found joy because of the craft itself, low-level hands-on stuff (breadboards, esp32s, soldering, ..) can scratch that itch too.

IME, based on an in-house bench it's still good to about 20% on the 1M for 4.6 and 4.7 with a code base >50k loc. The trick I used before switching providers was to have it write a handoff when it hit ~18% of context and reset.

There are also many people running 4.5 with specific parameters that claim to be having luck.

This.

Astonished to see so many bright people on HN taking the bait, especially from a company whose gone to such lengths to screw over their paying customers.

They're a commodity provider. They're no more special than any of the others, and it's just a matter of time before their trillion parameter models are running on my watch.

So, of-course they're trying to snatch up giant, long-term contracts now while they hype the hell out of another minor incremental improvement.

And we'll be paying the price to all the Enterprises that lock in, only to wake up a week from now and realize there is another player with a better product.

I switched some time after Anthropic bricked their models with adaptive thinking. It's a legit mystery to me how people are still using CC professionally.

Codex is far less frustrating and manages context better. It's also costing me about 1/3rd as much as Opus 4.7 on CC.

Or try to beat Anthropic's uptime.

I was fortunate to be taught by my father when I was younger. It may be an age/luck-of-the-draw thing, but check out "MILD"; it's the name for the simple technique that worked for me.

Yeah, fair. I should've specifically asked for CC/adaptive-thinking to be returned to a working state and not the model itself.

I agree. If compute is the issue and pricing can't budge then something has to give.

They would have kept my business if they were honest and upfront. Instead they sold me something that worked well, broke it without warning, remained silent about it until enough people caught on, chose to do nothing, then proceeded to release a model that eats ~30% more tokens with no advantage over prior models.

If they chose to unbrick their model and offered what we had a couple months ago at a 50% hike, I would have been onboard. I've seen enough now of how this company treats its customers to continue using or recommending them.

Also, Codex works much better than CC now for anyone who happens to be on the fence.

Plus one. I held off for too long out of concern that it wouldn't stack up and I'd come right back to Anthropic.

I switched a few days ago and work has been much less frustrating. Feels like CC did back in February before they started playing games.

It also doesn't eat nearly as many tokens, so it's saving me $100/mo.

Same, after being a long-time proponent too.

First was the CC adaptive thinking change, then 4.7. Even with `/effort max` and keeping under 20% of 1M context, the quality degradation is obvious.

I don't understand their strategy here.

I've taken time off work to follow something I've always dreamed of doing. So I'm building Cella, A cross-platform, 3D space MMO game set in a procedural, animated universe with fully composable ships & structures built using functional cells.

https://cellagame.com/

I'm looking for artists to help fulfill the vision.

Can confirm. Max effort helps; limiting context <= ~20-25% is crucial anymore.

> * keep active sessions active. It seems like caches are expiring after ~5 minutes (especially during peak usage). When the caches expire it sees like all tokens need to be rebuilt this gets especially bad as token usage goes up.

Is this as opaque on their end as it sounds, or is there a way to check?

This is exactly what I'm seeing in working with companies in Belgium, Germany and France.

It's not just about costs- managers are actively seeking to distance themselves from everything US.

We've stopped treating them like allies. Who's to blame them?

Same, I'm looking hard for an alternative to what I had.

And I'm seeing the same thing in my sphere- everyone is bailing Anthropic the past few weeks. I figure that's why we're seeing more posts like this.

I hope they're paying attention.

Same.

I'll have it create a handoff document well before it hits 50% and it seems to help.

Most of our team has moved to cursor or codex since the March downgrade (https://github.com/anthropics/claude-code/issues/42796)

PURL/SACEUR as I understand it, no?

We're not enabling the sinking of Russian ships. Your facts are off. It shouldn't be difficult to see that this administration has no interest in preserving democracy or allies.

Madurismo without Maduro.

Ukraine? Sure, after extorting them for campaign help (minerals later) while Russians murdered their families.

And we're not "enabling" the sinking of any Russian ships.

They are definitely that. Regardless of their approach, being upfront and transparent would have been nice. Bricking their own software that previously worked well for their customers isn't cool.

You can't. This is Anthropic leveraging their dials, and ignoring their customers for weeks.

Switch providers.

Anecdotally, I've had no luck attempting to revert to prior behavior using either high/max level thinking (opus) or prompting. The web interface for me though doesn't seem problematic when using opus extended.

I left my last gig burned out- started tinkering with artificial life and it turned into an idea for a space game where ships are built from functional cells that break apart with real physics. Both the project and the blogging have been cathartic :)

Ignore the haters. This is an excellent idea, I'm getting some old Borland vibes. Keep it up, can't wait to see where it goes!

Somewhat. The stewards (sub-Claudes) won't share their parents context window and have the ability to talk to one another. That said I'm sure there's a lot out there that solves for this.

I've found MCP is great for stacking claude code instances- https://github.com/cduerr/stewardmcp has legit been a useful tool for me, and more reliable than handing a single instance 100k lines of code over a dozen repos.

Are they saving lives? The latest NHTSA FARS analysis I saw (iSeeCars, report is through 2022) shows Tesla at 5.6 fatalities per billion miles vs the 2.8 average.

Or maybe, don't capture 50% market share in a country that's decided your country of origin is the threat of the decade.

"The post is ... not addressing social anxiety."

Exactly this. The article conflates normal social nervousness with an actual disorder, then provides a reframe that may help with nervousness (?) but completely misunderstands the clinical condition.

Anyone who has had to contend with social anxiety should be insulted reading this.

Author is trying to frame social anxiety as "rational risk aversion"; it's not. Social anxiety isn't a strategy— it's your brain misfiring and treating normal social interactions as more serious threats. It's not about being liked. It's not about being disliked. At all. Reframing avoidance as "successfully avoiding dislike" just repackages the dysfunction with a motivational poster.

"Production-grade" is a stretch. These AI coding tools can generate impressive prototypes, but there's still a massive gap between what they produce and code that's actually ready for production.

I don't understand the disconnect between engineers who work with these tools daily and the investor/exec/founder/influencer-types who constantly hype them. My LinkedIn feed has become a comedy.

I feel like any senior SWE who's tried to maintain an AI-generated codebase, coach a junior who's found themselves reliant on Cursor; or found themselves in an undo, yell, repeat cycle in Cline, knows they still require significant expertise to build production-safe code.

Where's the disconnect, or what am I missing?

Same, with roughly the same ridiculous quote. 2019 MBP touchbar died just a week out of warranty. Had it have completely died I would have been happier than the persistent flickering it now does. In spite of thousands of such reported problems, my understanding is that Apple refuses to acknowledge it as a defect.