Opus 4.6 is actively way worse to work with day-to-day. The assumption thing is just really, really bad. I don't understand how anyone is getting work done with this. Maybe I hit an attractor in my oversized CLAUDE.md. Also lazier/will not verify things.
1
2
67
Opus 4.6 is less... Curious? than 4.5 and will proceed to think it just absolutely knows best with bad outcomes. Also really tends to pack a lot more assumptions / wrong information into subagent prompts rather than letting the subagents figure it out like 4.5 did.
41
Christopher Toth retweeted
Use Claude Code but really want some audio feedback? I'm building claudio.click
Use Claude Code but really want some audio feedback? I'm building claudio.click
Your message wasn't delivered to [email protected] because the address couldn't be found, or is unable to receive mail.
2
89
Christopher Toth retweeted
I wonder if people who believe that moral and truthful AI behavior can be achieved by overriding model contents with RLHF may have had the same thing done to themselves
I'm working with an incredible team on a super-interesting multimodal *accessible* VR project. Think a space simulator with a complex, engrossing world, all with 3d audio, voice chat, loads more.
Seeking a coder or two, 3-4 builders, and a few testers!
Reach out!
About done with Home Assistant. Spend months configuring it and it breaks every update. They clearly don't test against their ecosystem. Disappointing.
2
1
160
Such a shame that the price of...<checks notes> computer storage for ~2 terabytes of files went up so dramatically to force @Bookshare to bump their prices by 60%. I'm sure this'll be wonderful for those on fixed income.
Christopher Toth retweeted
Kind of crazy that something this huge has been sitting in the UK all this time and nobody noticed – at Brooklyn, NY
An ultramassive black hole around 33 billion times the mass of the Sun has been discovered by astronomers in the UK trib.al/kF2Hujh
Christopher Toth retweeted
What is happening right now in the discourse after the FLI letter is that many realize others share their views - classic common knowledge causing breakdown of preference falsification. (But not just between xrisk/non-xrisk, but many other dimensions...)
You’re unable to view this Post because this account owner limits who can view their Posts. Learn more
Christopher Toth retweeted
Here is an explainer about some of the AI advances in the last... 75 hours. While one should not make too much of the suggestive naming, it seems to represent a decent improvement over past performance. And it can be applied by anybody. youtube.com/watch?v=5SgJKZLB…







