I'm James. Living between Beijing and London. I like coding. Also plants. Stroke survivor & disability advocate. My dog's a whippet/iggy cross and is called Ducky. He's a beautiful lunatic.
* website: [j11y.io](https://j11y.io)
* building: [nope](https://nope.net)
* bsky: [@j11y.io](https://bsky.app/profile/j11y.io)
* contact: https://tally.so/r/waYPvE
* twitter: [@padolsey](https://x.com/padolsey)
* book recommendations: [ablf.io](https://ablf.io)
* me = founding eng @ [collective intelligence project](https://cip.org)
My work email is my first name at cip dot org.
---
---
My meet.hn token: meet.hn-10d164d4-bf7f-4431-898f-e02b34b836dc
That's good. I wonder if it should be opt-in instead of opt-out. Disabled people are arguably less able to find random configuration options than non-disabled counterparts. I get a bit bothered with how undiscoverable these options are. But power-users by their nature don't mind going to the extra mile to get perf out of their experiences.
This is just the 'LLM judge', very badly implemented without any scientific prudence. What a joke. To be terse: you cannot rely on LLMs to provide standardized scores against arbitrary criteria. To get close to 'reliable' you would need highly tested rubrics, grounded in human decision-making, and you'd need to avoid all the measurement biases these things are riddled with... positional/order effects, anchoring on whatever numbers you stuffed into your own prompt, scale-format sensitivity (a 1–5 and an A–E scale give different answers for the same input), holistic-vs-isolated context effects, and lovely examples like where adding a "be unbiased" instruction makes it more biased. I've studied this at length. You cannot even _begin_ to approach this problem seriously without held-out validation, inter-rater agreement, and ground truth. This repo is just quagmire of wishful vibes with random numbers littered throughout.
These vast multi-agentic systems with roles like 'Researcher', 'Writer' (with a review loop), 'Reflection agent', seem to ~feel~ mostly right but lack evals as to the merit of agent decomposition. So it forms a satisfying enough flowchart but I see no evidence these authors actually tried other approaches or agent roles. And let's be honest: an agent is just a system prompt and output contracts, and these rich architectures seem to be pontificating beyond their worth. It all feels a bit vibe-y.
The most extreme virtue-signal is to go completely browser-default and have no styling whatsoever. Like lowercasing because your pinky can't be arsed to reach for the shift-key even though you've a billion dollars in series A.
>Where cash is stigmatized? I haven't seen such a place except PRC.
I don't know about stigma, per se, but there are a few places where businesses have a pretty explicit legal right to refuse cash, UK, Netherlands, Sweden, US. Oddly in PRC refusing cash is illegal.
Does anyone know how this even happens inside the walls of google? Is it an automated process? How is such a (presumably) high revenue account just magically blocked without human intervention? I'm quite perplexed.
My understanding was that strokes caused brain cell death, and that there was no coming back from that, but my neurologists would speak of 'bruised' brain cells, and that after weeks or months or even years you can see recovered function. UCLA's work here is targeting this disconnection and the lost rhythm in the surviving, distant networks. However there is, as yet, NO concievable intervention that could recover function from cell death at that center of the infarct.
This is fun. I'd like to see the same idea but oriented for richer tokens instead of simpler tokens. If you want to spend less tokens, then spend the 'good' ones. So, instead of saying 'make good' you could say 'improve idiomatically' or something. Depends on one's needs. I try to imagine every single token as an opportunity to bend/expand/limit the geometries I have access to. Language is a beautiful modulator to apply to reality, so I'll wager applying it with pedantic finesse will bring finer outputs than brutish humphs of cavemen. But let's see the benchmarks!
Knowing the nature of a test ahead of time, building out your capabilities and tooling before entering the exam hall when your peers don't have that advantage, makes you a cheater.
Does it have to be mutually exclusive? On-the-fly software does not destroy software. Gatekeeping software creation does not mean shoving the existing creators out, it just means creating a larger space that others can occupy, like when 'real' programmers had to slowly permit 'script kiddies' into their spaces. All feels a bit 'old guard' vs 'new guard'.
There is indeed a painful dissonance here. I like this new world, but feel sorrow for the loss of something. I try to remember how empowering AI is. It is already allowing millions of people to finally use the devices they've been sitting in front of all these years. No longer do they have to feel constrained by software creators who have made choices for them. Now it is their tool through-and-through, and they can construct software on-the-fly to match their needs precisely. They have been buying computers with both hands tied behind their backs. Now they are in control.
I'm James. Living between Beijing and London. I like coding. Also plants. Stroke survivor & disability advocate. My dog's a whippet/iggy cross and is called Ducky. He's a beautiful lunatic.
* website: [j11y.io](https://j11y.io) * building: [nope](https://nope.net) * bsky: [@j11y.io](https://bsky.app/profile/j11y.io) * contact: https://tally.so/r/waYPvE * twitter: [@padolsey](https://x.com/padolsey) * book recommendations: [ablf.io](https://ablf.io) * me = founding eng @ [collective intelligence project](https://cip.org)
My work email is my first name at cip dot org.
---
---
My meet.hn token: meet.hn-10d164d4-bf7f-4431-898f-e02b34b836dc