machiaweliczny

I think it's best ranking of models so far

But with good computers you can create simulation of biology and verify if via known and described effects. That's how it's gonna be done IMO

Anyone knows it Nautilus Biotechnology could really be this useful for decoding exactly that?

Man I would love the superhuman math AI when I was at uni, there were some problems I couldn't even grasp and this always bothered me.

I guess motivated state actors might still find a way to cause issues

My layman intuition is that given how easy is to make Turing machine and how good NNs are at fitting function I suspect they don't have hard time imitating whole logic via "stochastic process" and recent math results kinda hint at it. Animals are using more efficient algorithm, they infer better the causality but I thought that bayesian algorithm is stochastic.

Thinking has to generalise somehow (likely via internal world modeling) or it's just random search with better and better heuristics that you can verify with real word. The heuristic you get is likely why there might be such big differences between humans for example in abstract problems as for baseline survival we are kinda similar as it got pruned to baseline thing that works.

I think LLMs are missing explicit world generation to be "self-contained" but world models + LLMs are basically the same thing that allows this kind of bayesian causality discovery training but maybe less efficient.

Most people for example when thrown some weird before them like cow from tornado while driving will likely just freeze instead of nicely routing around it etc.

It likely hints that it's so much out of distribution that some heavy computational process is starting trying to model what happened or simply there's no good action to chose as you are "unprepared" for this distribution of event at all and likely both are true. I think that's why self driving cars also have problems with these weird edge cases as you need perfect word modeling to know WTF is happening in these edge cases and they simply won't so it has to be very simple heuristic like stop or try not hitting thing whatever it is in the end. We already have good modeling of human and animal behaviour built-in so driving is a bit simpler from computing these aspects at least but IMO perception is really underappreciated how hard it is but recent vision models seems to be approaching human ability.

If it's cheaper then I guess it will might get practical. Also more likely that mind uses simpler techniques more similar to these.

I wonder though if someone tested mutating training objective though, like keeping original loss/goal and somehow defining loss differently and then comparing against original. This intuitively feels like how mind tries to handle difficult tasks.

In general I like Svelte and SvelteKit although there's 3 problems I don't fully like:

1) The getter trap - if you forget to export hooks via getters you don't get reactivity and it's not easy to figure out (Svelte)

2) The async data fetching - for example fetching data async is currently coupled to views which it shouldn't be - was frustrating a lot (SvelteKit)

3) That reactivity system is not runtime only (Svelte)

I am also not a fan of adding these new things like remote functions - most of apps need multiple frontends and this coupling isn't needed. This will be fad same as GraphQL

Agree, with UA stopping Russia western capital will finally flow into Poland because previous strategy was to defend on Vistula river which blocked a lot of heavy industry investments that are starting now.

> It is horrible until AI gets as good as a perfect project from a prompt or we need to abandon it.

Getting close to it with my recent test of cursor cloud workflow.

I've spent 300M tokens in a day and it achieved what I wanted. Manually would take me a month instead of day so it's not 2x it's 20x faster but project was quite simple rewrite of 20k lines of C++ and there was reference implementation. It also improved on original implementation perf wise.

The problem is with fuzzy ideas for new development and with catching up with comprehension when working on something new - I guess we need better visualization tools for code. Product manager work seems like bottleneck currently and monitoring because coding can move quite fast.

Good news is that HTML in canvas might bring back these cool days :)

Yes it's still bad there's no viable headless UI in browser one can really style and it has all the a11y etc. but need extra library for selects that work etc. Invented work for no good reason. The real complexity is in diversity of devices though nowadays in the frontend.

At current rates you are losing when renting (excluding possible appreciation), that's why it's hard to sell and easy to rent. Due to low rates before homes went big in value cause you were earning even not counting appreciation, so whoever had credit was maxing it to full for easy money but now situation starts to stabilise / reverse?

The thing with home is that it's leveraged investment for normal people. You can pay 10% of value and resell later. Buying home for cash is less optimal but for credit it makes sense usually due to leverage and little downside risk. Plus usually every now and then you get money printing and you 2x your leveraged money or more rarely get hyperinflation and home for free. This was mostly issue of too low interest rates but that's how I see it from the past 10 years (non US resident)

Tested and it's good. Fast version is bad though. I like planning model in Cursor that it works more like human written design doc instead of too detailed AI plan. Seems like this is more responsible for results that model but still on fast it failed but on normal got good results.

The practical case is less time spend on rebasing/formatting code - IMO formatting standard is very helpful as then it's the same as storing AST. There's also better preserving of git blame and that's likely why they have done it as single operation as otherwise you would have everybody messing with that part and now you know there's single commit that touched everything and if blame is on it then you check previous edit.

They would compete with Meta then which already does something like this but has mature AD tech

Rookie mistake, you just conflating concepts pretending things don't exist. Good API is one that is minimal, explicit and acomplishes the job without conflating with somethings else. Congrats now I don't know if this variable is designed for rendering of state managements and when I will hit some perf issues will have no idea what causes this, same with some unintended state update or something like this. If you can do everything using same concept it's obviously harder to read as you need to simulate compiler in your head which is bad design.

I personally go 3x for gym classes like "healthy spine", "mobility", "core" etc. and then 2-3x hard training. But I would say I am very active recently.

Do I feel better? Yes. Was it hard first 2 weeks? Yes, I had even to resort to painkillers.

I think the best for people who sit a lot are core, mobility and back exercises. Huge motivation for me when I finally started prioritizing back on machines and progressed on all other things and finally look like I go to gym :)

5.2 and 5.3 are strong/best for coding, 5.0 and 5.1 were garbage

We need that for this chinese 3B model that think 45s for hello world but also solves math.

A lot of NLP tasks could benefit from this

No it's not because cost is much lower. They do some kind of speculative decoding in monte-carlo way If I had to guess as humans do it this way is my hunch. What I mean it's kinda the way you describe but much more efficient.

Windows Phone was actually good. I would even say that my Lumia something was one of best experiences ever on mobile. G+ was also good. Efficient markets mean that you can "extract" rent, via selling data or attention etc. not realy what is good

--help explains this, -s for sandbox and -a for approvals

For now until they are in the lead

Claude: you get rate-limited with one prompt so hard to validate 4.6

Codex: better with rate-limits, 5.2 strong with logic problems

Cursor: cursor auto - a bit dumb still but I use the most for writing not really thinking, it's also good at searching through codebase and doing summaries etc.

Claude / Codex still miss tons of scaffolding for sane development or it's due to sandboxes or sth. Like for example you ask in /plan mode to check think with link to github and it does navigate github via curl, hitting rate limits etc. instead of just git clone, repomix etc. so scaffolding still matters a lot. Like it still lacks a tons of common sense

I think this will work but requires quality data pipelines and scaffolding same as coding

Yeah, I think people have had enough this time its not gonna fly

Man this is art