bret

progressive plus preserver, p-table walker, euphemia predictor,

[bret @BretHatin]

Weird

[Kaarssteun @kaarssteun]

@Xinyu2ML the full graph makes this even weirder

[bret @BretHatin]

Yeah this just reads like rehashed applied math research vs foundational math research, and i think you overestimate the abilities of humans to immediately know the point of fundamental research, when time and again we have realized its application only after decades. Its a lazy argument, is all.

[bret @BretHatin]

@mitchellh Unfortunate....simd post was tuff tho

[bret @BretHatin]

This is unsurprising because they are Germans, and Germans unfortunately are famous to take ideology to their maximum.

[Armin Ronacher ⇌ @mitsuhiko]

I think this is a very bad move, and the people behind Codeberg should re-consider there stance. https://t.co/mTsq4K4Hso

[bret @BretHatin]

If you are letting AI run through your notes, you deserve to be in permanent underclass.

[zoryn @z0rynx]

A 27-year-old pointed Claude Code at 4,000 Obsidian notes and woke up to an employee he never hired.

6 years of notes. Ideas, half-finished essays, 212 book summaries all rotting in a folder he stopped opening in 2024.

Then he noticed something obvious that almost nobody uses.

Obsidian is just markdown files, and Claude Code lives in the terminal which means his entire second brain is one cd command away from an AI that can read all of it.

So he ran it once, overnight.

By 7 AM the agent had crawled every note, linked 340 orphan ideas back into the graph, flagged 18 contradictions between things he believed in 2021 and things he wrote in 2025, and drafted 6 essays from threads he forgot existed.

Now the loop runs every night while he sleeps.

New note goes in messy wakes up tagged, linked, connected to 3 older notes he'd never have found himself. His daily note greets him with what he was thinking exactly 1 year ago and why he was wrong.

Cost of the setup: 1 folder path and a 400-word CLAUDE.md file.

No plugins, no Notion AI subscription, no $30/month "second brain" course.

People spent a decade building vaults that just sit there.

He built one that thinks back.

[bret @BretHatin]

E = mc^2 + AI

[Graham Christensen @grhmc]

🙏 P = NP + AI 🙏

[bret @BretHatin]

Ever since i have been using LLMs for work, 99% of menial work has been automated and I have been exclusively working on only hard problems. It has encouraged me to work more because i find meaning in working on hard problems. I go home a lot more tired now than 6 months ago because I do alot more in a day than I could have done earlier.

[Isaac King 🔎 @IsaacKing314]

I don't fully understand why programmers seem to be the ~only group of people not angry that their jobs are being automated.

[bret @BretHatin]

hedonic trademill

[bret @BretHatin]

How can you be this stupid?

[Файн @Perfectly_Fine1]

@LegacyMediaFail Idk, fucking google it. The idea comes from observing brown water in a tap one day.

[Kimi.ai @Kimi_Moonshot]

Self-evolving: AttnRes Kernel Optimization

Given FLA Triton AttnRes at production scale (96 layers, 8192-dim model, 8192 tokens), the goal was to maximize training-side speed without changing numerics.

Over 15 hours of nonstop iteration, K3 designed a novel two-phase kernel algorithm, fused kernels while preserving numerics, and reduced forward+backward time from 283.6 ms to 114.4 ms.

K3 and Fable-5 (with potential fallback) reached similar performance, but K3 improved faster per iteration.

[bret @BretHatin]

oh we cooked cooked

[Angel 🌼 @Angaisb_]

I thought open-source chinese models were supposed to be cheap...

[bret @BretHatin]

oh we cooked cooked

[Hensen Juang @basedjensen]

JFC if this is true then yea china bros are hot on the heals of murican frontier bros and scale is all you need

[bret @BretHatin]

Nooooo, you dont understand you have to do a humanities degree and have to spend 5 years typing word documents for a living to talk policy with us....wait are you saying what your model is telling you to say!?

[Pieter Garicano @pietergaricano]

Unfortunately "Member of Technical Staff" did not resonate with European Parliamentarians

[bret @BretHatin]

Man it kinda feels like we are bottle necked by our memories and our abilities to read stuff slowly. I wish to remember alot of stuff and read a things very quickly.

[bret @BretHatin]

Are we finally back?

[0xmer @WareTheWock]

0xmer space Sunday yay or nay

[bret @BretHatin]

Berlin also has this and you can use that data to find pairs of tree that are perfect to hang a hammock on.

[Benjamin (offline for a bit) @bschne]

TIL Zurich keeps a numbered inventory of all trees on public grounds, which is accessible online and can be downloaded in various GIS formats

[bret @BretHatin]

we are circling back to pre 2025 discussions here about how it is impossible to work with arbitrary file sizes unless you use GC or RAII #noticing

[hardfist @hardfist_1]

I like Zig too, but I’m not convinced by the argument that TigerBeetle proves memory management is not hard in Zig, and that projects struggling with it simply have a “skill issue.”

TigerBeetle’s memory model is fundamentally different from that of a typical application. Many real-world workloads are difficult, or even impossible, to estimate in advance. For something like the Bun bundler(or Rspack), which has to process arbitrary user code, I don’t see how you could reliably predict an upper bound on memory usage ahead of time.

So we may need a better example(ghostty maybe a better one) of how memory can be managed well in a large, general-purpose system—especially in complex scenarios involving FFI, like Bun. TigerBeetle is an impressive project, but it is not a representative case.

[bret @BretHatin]

kinda crazy that you could just raw dog kernels on AMD by just ioctling the public kernel api

[bret @BretHatin]

@seatedro Very smug but coping vibes