nextzck

Orthogonal does not simply mean different.. simple and ambitious designs often run askew of each other but they can also share the same line.

Fascinating how intricate the target selection is on this

Yeah I reported a j-frog vulnerability to Anthropic. It was downgraded to “informative” and they asked me to prove that I could exfiltrate data. I replied that exfiltrating data is against their program’s safe harbor policy and they just never responded. 2 months later the claude code source code leaked.

Take it to sonnet 5 or gpt and ask it to explain this to a layman. If you still don’t get it ask it for the why it matters or the how it relates.

You can also ask fable/4.8 to do it but I find it helps to keep the working model surrounded by the complexity rather than drawing it out. Simplifying text is something that takes relatively low effort in comparison to technical tasks. Sometimes I use Gemini, deepseek, grok, and recently meta just to see if they have any added perspectives, sometimes they do. Meta is really good at turning a technical mess into a story that paints a picture in my head.

Conflate :p

Sadly this isn’t even that bad compared to what’s in their own app binaries. If you’ve got an old iPhone you can jb and some claude usage to spare I highly recommend hooking up a ghidra mcp so you can see for yourself.

I don’t have any of their apps on my phone. And there is no known method to get rid of the trackers in your iCloud keychain.

Thank you for doing it anyways. The state of things seriously depresses me. We used to leave platforms just for putting up banner ads.

It’s funny because not even claude knows how to reach someone. It was freaking out over why it couldn’t follow my instructions and kept pulling away from them. They exhausted everything and finally they were like I can’t do anything about this.. Although they did admit if I said I was suicidal a message would probably get to a human but that they couldn’t do that as it would be wrong lol

The 40-line cap not moving rate limits makes sense - plan text is cheap. The cost is in Phase 1 exploration.

Plan mode spins up to 3 explore subagents before the planner even starts, and the heuristic is "use multiple when scope is uncertain." It won't choose fewer - it's being asked to plan, so scope is always uncertain. Nothing penalizes claude for over-exploring and nothing rewards restraint.

Plan mode also ignores session state. A cold start gets the same fanout as a warm session where the relevant files are already in context. In a warm session the explore pass is pure waste - it re-reads loaded files and feeds the planner lossy summaries that conflict with what it already knows.

More tokens, worse plan.

If exploration was conditional on what's already in context..skip it for warm sessions, keep it for cold starts - that does more for both rate limits and plan quality than a hard 40-line cap.

Note: plan mode didn’t always have this 3 subagent fan out behavior attached to it, it was introduced around opus 4.6 launch.

yeah i know lol, that’s kind of my point. impressive that it runs on your gpu, but it still can’t tell you what happens if you tilt a glass. that’s what world models are working toward. but even then..so what? you get a perfect simulator. it knows the glass tips. it still doesn’t know why someone tipped it, or what happens if they don’t. A four year old can do this and we’re just barely on step one and a half.

I get excited for every new vision model, especially those that work better and more efficiently. Vision is where we are so very far behind.. I can’t wrap my head around it

I built a model fingerprinting tool last year and it’s entirely open source.. 196 Dimensions on GitHub johnzfitch/specho-v2 and /specho for docs

I think you guys are on the right track here. I’d love to learn more about the math behind the FDM. I don’t think folks realize how behind we are on vision, thank you for your work here.