clord

I don’t really use off the shelf agents. Usually just string together what I need. In this case I give the lead agent an mcp tool and tell it to use the service to track all work, and what should count and not. For work I use private GitHub project board and for personal stuff obsidian and markdown.

I don’t like big fat agents that try to do too much. Better to keep it lean, each agent prompt is really just a short brief, like a paragraph or two. If it’s longer consider breaking off skills and tell the agent the list of skills it might use.

I notice I am doing more work as well but enjoying it and taking more time to recharge. I stopped listening to podcasts and audio books while working because before, my linguistic brain had time to linger while puzzling. Now I’m reading and writing letters and memos all day in addition to that other puzzle solving mode. No mental bandwidth for those extra things.

And multitasking is easier with ai. Have a lead agent keep track of your work. With a real task tool. I can drive several projects at once. I finally (have an agent) keep my work journal updated.

I do miss raw coding sometimes. But if I am honest most of what I did was filling in boilerplate code that was not really new or interesting. The real thinking was not the code itself. The code was a means.

I used to worry about who will learn coding anymore but seeing how I can push into new types of systems now, my worry is reduced. Kids will just have really awesome assistants who can explain it all in the way they prefer.

They can. I had one doing original OCR of a scanned piece yesterday. It found relevant page numbers and understood the context. It read the whole book to find the relevant cites. Even went though the genealogy table and found ground breaking references no one noticed before. Total cost under 5 dollars and time was less than 20 mins.

The hallucination era is rapidly ending.

Very insightful actually. Taste and vision are both instances of holding a model that predicts a good result beyond the threshold of validity in theory but reality happens to align. Is it luck then? Perhaps meta-luck where lucky weights produce “accidentally great” results with some predictability.

Steve Jobs had a mental model that brought the iPhone. No one really wanted it but something in his life biased the result.

So creativity is having weights so good you can project way out into latent space beyond what is reasonable.

I suppose the Georgian calendar includes this gap officially? Or did the catholic world just jump from system to system? If it includes the gap officially, then yea, error is the right thing. But if we just moved to a new calendar, then arguably there should be no error. both calendars work for all dates continuously, and we could have conversion functions etc. The gap is just a big flag day (week.)

imo, if i am using Georgian i want that system even retroactively, just like projecting a date back into the stone age. I don't care that a particular geography or culture changed calendars.

So I suppose what is really going on is that some libraries are using Georgian consistently, and others are using some hybrid "civic" calendar that tracks what the romans and the catholics were using, so dates in history books make sense.

The intention is to make editing easy and quick on slow and memory deficient computers. This is how for example editing a pdf with form field values can be so fast. It’s just appending new values for those nodes. If you need to omit edits you’d have to regenerate a fresh pdf from the root.

There is something deep in this observation. When I reflect on how I write code, sometimes it’s backwards. Sometimes I start with the data and work back through to the outer functions, unnesting as I go. Sometimes I start with the final return and work back to the inputs. I notice sometimes LLMs should work this way, but can’t. So they end up rewriting from the start.

Makes me wonder if future llms will be composing nonlinear things and be able to work in non-token-order spaces temporarily, or will have a way to map their output back to linear token order. I know nonlinear thinking is common while writing code though. current llms might be hiding a deficit by having a large and perfect context window.

I think that whole section of the post was "For example, in Kotlin," presumably these are true in that language.

I was under the impression these offices closed during the pandemic and the return to office order is bringing people back into those places. If that’s the case I don’t think this is some sort of planned disruption but rather poor planning. Incompetence vs malice right?

The automation should be setting flags on videos. Users should have preferences for opting in or out of flags with reasonable defaults. If there is a jurisdictional requirement in a users location YouTube sets the preference to disabled according to the law and shows a link to the regional law so users understand.

Hence abuse is a local thing too. One can be getting flagged in one region but not in another. ‘Abuse’ amounts to getting certain flags auto-applied in some locations or whatever. Should not affect the account itself though.

I have a planned trip to work around. I want to make sure those are booked, and allocate the rest optimally. I suppose this is the same problem as having extra days to allocate.

I used to do something like this all the time with C/C++ compiler tests. I tried lots of fancy tools and stuff, but I kept going back to: expand all macros and retokenize to one token per line (I made a custom build of the preprocessor that had this built in). Then, have a shell script randomly remove lines, and use another script to check that the resulting test case behaves consistently with the failure. It would run for a few hours (or days, for boost problems), then usually you'd get a really minimal testcase that shows the problem. Often I would use this to find regressions. Just have the shell script check one is good, the other has the problem. The resulting output would then usually point exactly at the regressed feature, and make an amazing unit test.

Before leaving compiler team, I wanted to find a way to do this one the AST level (so parens always go in pairs, etc), but that could also complect with the bug.

I wonder if LLMs could accelerate this by more intelligently removing stuff first, iteratively?

I used chrome to confirm whether my DOM was upgraded, and saw the same result. The graphs were very synched up.