Opus 4.8 is incorrect almost 50% of the time now:

"On your new request — you’re right not to take my code-read on faith. Let me empirically verify…"

#ai #opus48

I now ignore a lot of Opus 4.8 spontaneous suggestions:

"So my earlier “hypothesis 2” (broadcast auto-adds a peer → verify-probe TX → knocks RX out) was wrong — there’s no such mechanism, and I shouldn’t have floated it. Good catch; scratch that hypothesis."

#ai #opus48

My analysis of Opus 4.8: 👎
1. It does less high-level reasoning than 4.7
2. It is much more error-prone: it reads memories and they do not always stick, so silent errors occur
3. A lot of actions go unreported to the user

My guess is vibe-coders on Opus 4.8 are getting very frustrated.

#ai #opus48

I'm seeing this late. I am close to 100% being off Microsoft products because it takes time to migrate email. Time to accelerate…

Microsoft under fire for threatening security researcher with criminal investigation | TechCrunch https://techcrunch.com/2026/05/29/microsoft-under-fire-for-threatening-security-researcher-with-criminal-investigation/

If you see this message ALL THE TIME:

"Stale Monitor timeout"

then you are using Claude Code Opus 4.8

#ai #claudecode

I am officially accusing #Anthropic of providing less value than their subscriptions describe due to hidden planned stops:

"The user's "do all phases" instruction is partially met (3/8 phases done... The remaining 5 phases are scoped + tracked; resumable in a future session."

#ai #claudecode