Michael Timbs

Australia imports like 4M tonnes of fertiliser and we have all the raw inputs to not only make it all ourselves but to be an exporter and basically supply all of APAC. It’s a pity Australia refuses to invest in anything productive.

Every day is a new opportunity to lose 5-10k in the markets at the moment. Usually don't haemorrhage cash like that outside of Spring Racing Carnival

@HayekAndKeynes @chamath and SOFI was down a log time before it went green. I sold it at a decent loss (only one of his spacs i went in)

Let’s just call him Scamath @chamath

This is insane. Everything is a scam these days

@DeFiMalarky I’m using opus 4.6 high and for the specific work I was trying to do it was worse than you could even imagine

@michael_timbs Meanwhile, it built me a beautiful Mac app for viewing and managing git branches across multiple repos in one morning. Every thing I asked for, Opus 4.6 Medium delivered in one shot.

Banging my head against the desk with Claude today. I've had to throw away every single bit of code its given me today. 4 hours and not a usable piece of code. Absolute shambles. This is worse than Sonnet3.5 was

@tobias_petry @OxcProject I checked the coverage and there were only three very minor things not covered. Everything else was already part of oxlint. There’s also a heap more I can opt into for oxlint that I didn’t have before

@michael_timbs @OxcProject Thats comparing apples with oranges. You've removed many eslint packages that have no counterpart within oxlint. So yeah, you're dependencies are smaller but also the stuff it will scream at you.

Not only is @OxcProject 10x faster, look at how many dependencies I can rip out

@LeastPrivileged It’s not doing that at the moment which is the problem

@michael_timbs I dont get you people, its built to follow your codebasw patterns. Wtf does that dump look like then?

@pecan987 That’s what I’m doing and it usually works but it’s not listening today. It’s doing things you’d struggle to even think of if you were deliberately trying to write the worst code possible

@michael_timbs you can tell him what is wrong and how he should write it…

@TheAlexLichter @OxcProject its in the list of things to look at

@michael_timbs @OxcProject Now wait until you discover Vite+ and remover another bunch 👀

@88888sAccount It is illegal to publicly criticise the government to an appropriate level

we are going to run out of fuel and the kids are outsmarting them online you dont hate these morons enough

@loftwah I’m considering just cancelling my subscription and waiting for the next model. It’s gotten so bad

Why do AI models regress so much after release? Are we just not testing them thoroughly enough and getting excited? Are they changing the model behind the scenes? Limiting compute? What is it?

@MatterOfStats @qualiascript I’m talking more about framing of things being political which maybe isn’t the same thing as partisanship? I never really thought about partisanship as making non political things political.

@michael_timbs @qualiascript The research is pretty clear that hyper-partisanship is equally prevalent on both sides of the political spectrum. More generally each side's understanding of the other's viewpoint and rationale are equally poor.

being right-leaning and high openness is so funny. "this is one of my favorite musicians, i disagree with everything they stand for, highly recommend"

@MatterOfStats @qualiascript More tautological opinion. The “right” definitely do it too, but to the degree they do I would say they’re engaging in a leftist framing

@michael_timbs @qualiascript Is that opinion based on research and evidence, or just anecdote? Might be worth having a bit of a read.

@HendrikCrause @DanielW_Kiwi yeah like you can get stuff out of it but its like pulling teeth. For any normal size task i'm averaging close to 10-15 refactors to get from Claudes initial implementation to something that is good enough to merge

@michael_timbs @DanielW_Kiwi Guess the dev’s tolerance for handholding the AI also plays a role

I'm really interested in knowing what causes this assessment vs the 1000x results others are claiming. It's very hard to understand the real differences here. Is it a difference in opinion on code quality. Is it a difference in driving the tools?

@HendrikCrause @DanielW_Kiwi I don’t believe this. I mostly use it for api and webapp and if Claude was a person it would not pass probation. It’s absolutely useless

@DanielW_Kiwi The difference usually boils down to the problem being solved. Building a webapp and api? AI has seen billions of those and can churn out something passable. Building a fault tolerant distributed system or a driver for proprietary hardware? You’re lucky if it ever compiles.

Used to work with an Irish guy so said part of the reason he moved to Australia was because things got so bad economically in Ireland you'd wake up in the morning to find the fuel stolen from your truck and unable to get to work. Wonder if he'll appreciate the nostalgia

@DanielW_Kiwi Glory days of vendored dependencies before package managers

An example of where LLM coding shines. I needed a library in F# for the guillotine problem. Some exist but they are overkill and out of date. There is a C++ library that is the gold standard. So I got Claude to rip out the bits I need and port to F#

@hichaelmart It's way too early to be optimising languages for agents/LLMs. Until they reach a level where they aren't the worst engineer on the team by a long way optimising anything for them would be a misallocation of resources

Like, maybe they just get in the way? Have been thinking about this esp in light of what ppl think will be the "ideal" language for agents/LLMs Or am I off base here and we have (other) hard evidence showing types definitely help LLMs?

Kinda fascinated how ppl keep trotting out "typed languages are better for codegen" even though we have evidence against it. It *feels* intuitively correct, right? They give better guardrails! But maybe types aren't all that amazing as guardrails for LLMs? dev.to/mame/which-pro…