chillacy

As I work with Claude more and gain a feel for its capabilities, I tend to run into 2 far less often, as I'll decompose my messages more for the current model limitations. The threshold also changes each release.

I think a reading of Roman history shows the failure modes of a senate so often that I wonder if it was ever supposed to work more than a few hundred years.

This looks more like "Human working at airline accidentally pasted their copilot prompt into whatsapp"

I switched from kimi to GLM for claws. I found it tends to be a stronger debugger when it comes to fixing openclaw issues with bash, being more thorough in finding the right root cause. Kimi tended to confidently misdiagnose. It's also much faster. The only downside is that it seems to cost an order of magnitude more than kimi.

A classic but I think relevant pg essay - https://paulgraham.com/makersschedule.html

The closest I attained to flow state vibe coding involved building UIs with claude 4.6 fast mode low thinking, a local TTS model (nvidia's), and hot reloading browser.

Arguably maybe I could have done better on higher thinking and then done more in parallel but it is more tiring.

If SWE Bench is public then Anthropic is at a minimum probably also looking at their SWE bench scores when making changes, I'd trust more a tracker which runs a private benchmark not known to Anthropic.

Yes, sometimes I have found it hard to sleep if I'm close to building something I want to build.

I think there is kind of a meme going around about multitaskers doing very well with vibe coding, and I can see it. Although, as someone who has the opposite problem, it can be tiring if I try to do more than two things at once.

I've found this to be true so far, junior engineers with AI can be super productive but they can also cause a lot of damage (more outages than ever) and AI amplifies the sometimes poorly designed code they can generate.

I suspect a lot of it best practices will be enforcing best practices via agents.md/claude.md to create a more constrained environment.

My take: openclaw should not run on a mac (even though looking at the skills it ships with it clearly was made to)

It should run on its own VPS with full root access, given api keys which have spending limits, given no way for strangers to talk to it. I treat it as a digital assistant, a separate entity, who may at some point decide to sell me out as any human stranger might, and then share personal info under that premise.

Check out https://x.com/jianxliao/status/2020667822800818253?s=46

Just uses claude. I haven't tried it much but it seems to be what you're describing.

Openclaw uses pi agent under the hood. Arguably most of the codebase could be replaced by systemd if you're running on a VPS for scheduling though, and then its a series of prompts on top of pi agent.

Pi's great. I really noticed it when trying some of the openclaw clones which try to be smaller in binary size and end up not using pi.

Boris has been very open about the 100% AI code writing rate and my own experience matches. If you have a typescript or common codebase, once you set your processes up correctly (you have tests / verification, you have a CLAUDE or AGENTS.md that you always add learnings to, you add skill files as you find repeatable tasks, you have automated code review), its not hard to achieve this.

Then the human touch points become coming up with what to build, reviewing the eng plans of the AI, and increasingly light code review of the underlying code, focusing on overall architectural decisions and only occasionally intervening to clean things up (again with AI)

Singapore’s free speech restrictions, whatever you think of them, no longer seem so far outside the box. Trump is suing plenty of people. The UK is sending police to knock on people’s doors for social media posts, and so on. That too makes Singapore more of a “normal country"

That seems like it should make Singapore _more_ cool, at least my personal theory is that this changed a lot of perception of China (at least in some parts of gen z social media, "it's a very Chinese time").

Korea has a similar demographic shape, and Japan already passed its peak in 2005ish https://www.populationpyramid.net/japan/2024/

I heard it's because the labs fine tune their models for their own harness. Same reason why claude does better in claude code than cursor.

Hmm, I once transited in Heathrow in a return flight from europe to the US and had to go through Heathrow security for whatever reason, where they subjected me to liquids rules way stricter than either my source or destination did.

E.g. 1 day use contact lenses and prescription creams all having to fit in a tiny plastic bag. So I'm happy for this change.

You can make still money off untargeted ads (just not as much).

Afaik the apps don't have to ask you, they could just request the OS-level permissions. They don't do that because if you reject the request at the OS level, they can't request it again, you have to go to the Settings app to enable it and it's harder to do. So apps prefer to just nag you again and again until you say you're ready.

Also not the easy kind of 1990's LED soldering, like tiny surface mount soldering.

2M units globally, so maybe generously 1M in the US (assuming you're from there but multiply by some factor proportional to your country's consumeristic tendencies), divided by the population is only something like a .3% rate of ownership. So not quite as prevalent as gopro, which has sold something like 35M in the US over the past 10 years [1]

[1] https://www.statista.com/statistics/688306/number-of-gopro-u...

The fiber is afaik a big factor for slowing rate and amount of sugar absorbed (amount because apparently some of it makes it far enough to feed gut bacteria in the large intestine).

Ironic that the NYT's article here focuses on the political angle instead of just the "facts" so to speak...

It is likely to please President-elect Trump and his allies.

The idea of some kind of universal fact is also misleading, some statements of fact are only statements of belief, others are so ill-defined that people end up debating two different things.

I think prediction markets (polymarket et al) get this right. Every question as vague as "is the earth warming" has resolution details which define some way to resolve the question such that all parties (even those with economic interest to disagree) have trouble disputing the outcome.

For a question like the earth warming, it would usually be something like "according to ___.org website on Y date", which in that case the final prediction becomes: will the average temperature in the period from 2016-2026 be greater than Y on ___.org, which is a bit different than the original but easier to arbitrate.

In case anyone else finds this interesting, the DOD has night vision goggles which have nearly-zero latency (all analog), amplify much better than digital cameras, and emit very little external light (hard to spot by adversaries).

This veritasium video at 12:00 shows how these goggles work: https://www.youtube.com/watch?v=UAeJHAFjwPM

That said I wouldn't be surprised if someone in the DOD was buying them for general tech exploration though.

Show us a link to a Facebook advertising API which can give me bidstream data which includes a user's demographic info? The links describe "Modern programmatic advertising" including "Patternz" but have not shown that Facebook is an implementor.

For the record every time I have advertised on both Google and Facebook, it works in reverse: I give them my ad, audience info, and how much I'm willing to bid, and they run an auction per impression (of which I don't get to participate in knowing the details directly). I get no user data, only the bill. If I did get user data (presumably with user consent) I would start my own direct-marketing mailing or email list and stop paying the ad-tech companies each time (who act as gate-keepers).

Hmm to what extent is there harm done to others in the form of increased future healthcare expenses if this is the UK we’re talking about?

Surprisingly similar to the results of humans in split brain studies, the brain seems to provide and accept pretty nonsensical explanations for behavior.

In my observation the people most excited aren’t writing top tier fiction, they’re writing marketing copy, SEO blog spam, and business emails.

But who knows, if this technology improves as fast as AlphaGo did, it may start taking on the best of the best one day?

One interesting way I heard to around this is by mixing human languages in the prompt which probably never appear together in any training data, and seeing that chat gpt can do still output sensible replies. That seems to imply that something unique is happening beyond token lookup, if it’s taking different languages and mapping that to the underlying information, that looks a lot more like what people call “understanding”.