Halvar Flake

I will have to print & frame this tweet :-))

@haroonmeer @PeteMarkowsky Perhaps the good variant of imposter syndrome thinking is: Feeling like you *could* achieve equality if you apply yourself, even if you're not there yet?

@halvarflake @PeteMarkowsky Yeah. Even if not the original intention, it nudges ppl towards “the delta is just in your head”. The general take is old/matches the Bukowski quote about intelligence/self-doubts

I think one of the most subtly damaging notions that was popularized over the past decade, was that of “imposter syndrome”. Lots of people quote “imposter syndrome” and “dunning-kruger” - but we’d be better off if we spotted our occurrences of the latter more the former.

@haroonmeer @PeteMarkowsky Ok, so what you're saying is: People say "I have imposter syndrome" to then stop investing in improvement? (Trying to understand)

@halvarflake @PeteMarkowsky It’s far less common.. I say: have heroes/role-models.. work till they know your work too

83% of my NYC infra founder buddies now acquired within a few months businesswire.com/news/home/2026…

@msuiche Perhaps. Perhaps. Have you tried telling Fable to audit the code it has generated?

I'm the opposite, I believe there will be a near future where models will write code with no bugs so appsec will die

@haroonmeer @PeteMarkowsky Have we? I sure still have that...

@PeteMarkowsky I gave a talk a while back where i mentioned that its not imposter syndrome when i stand next to @halvarflake or Dug Song. They _have_ done things i wish to. I feel like we've taken away "i want to be like X & am not yet (so will keep working at it)" & we are poorer for it

The amount of bugs and terrible mistakes that Fable will turn up in Opus-generated code is genuinely hilarious and makes me very hopeful for the future of vuln-dev.

I am intrigued that I agree with Geohot on so many things these days.

Build a compiler to solve Anthropic's interview challenge. @AnthropicAI published their take-home interview challenge on Feb/2026: optimize a kernel running on a simulated VLIW SIMD virtual machine to minimize cycle count for a tree traversal + hash computation workload. It has now been about 5 months since the challenge was released, so I am sharing my work as well. I hope this does not interfere with Anthropic’s interview process. The original intent of the challenge was for candidates to collaborate with an LLM and manually optimize the assembly program. I took a somewhat different approach: I built an optimizing compiler for the VLIW machine. The compiler can optimize not only the tree-traversal and hash program used in the interview challenge, but also the performance of arbitrary programs targeting this machine. My solution does not use any algorithm-level optimizations, nor does it contain any optimizations specifically tailored to this program or benchmark. Every optimization is a general-purpose compiler optimization. Result: CYCLES: 1137 Speedup over baseline: 129.93315743183817 Check it out if you are interested in techniques: Loop unrolling; Common subexpression elimination; Dead code elimination; Peephole optimizations; Scalar replacement of aggregate; Arry scalarization; Alias analysis; Memory dependency based load/store elimination; Straight line strength reduction; Data dependency analysis; SLP vectorization; Load-store vectorization; VLIW instruction scheduling; Register allocation; etc. github.com/fiigii/ai-comp

@teortaxesTex It sounds unbelievable.

> 12,000 Wh/kg theoretical limit lithium-air EV battery Insane If they get even 50% of the way there, The Electrostate model simply wins. All-electric logistics. Robots with week-long autonomy. Other storage types filling temporal/spatial gaps. Hyper-resilient grids. Near-utopia.

I think the impact of this document has not yet been understood. This is not some silly mumbling about "application scenarios". Jie Tang is a Tsinghua professor, one of the patriarchs of Chinese AI, leader of the GLM project. He says: yes, I know what AGI means. It must be open.

every research team needs to spend some time learning how their modeling code lowers down to gpu kernels example: rms norm vs layernorm industry assumes rms norm is cheaper. you don't need the x̂, it looks simpler, so it must be faster, so it got adopted but that's not true. both rms and layernorm are memory bound kernels (the amount of time it takes to get the data to gpu cores is longer than the amount of time it takes to do the computation on said cores) both take the same amount of time e2e so can probably train the model right with either, but maybe it makes a difference (e.g why diffusion models still use norms with an affine shift e.g adaLN) i get why. as open-source models get better and, inevitably, commoditized across the inference providers, overall serving speed determines user experience determines which model gets adopted but there's no reason to superstitiously avoid free things

3. as an employee in the right company, strive to be excellent, share freely, admit mistakes, "do the right thing". Because a good company rewards this behavior. The end. 17/

The probability the project will be of low quality is H^x * L^(N-x)=L^N * (H/L)^x. The impact is not linear in x/N, the "talent density". 12/

Because technology, organizations, and global reach of current markets multiply the impact of what we do. Code performance and effectiveness multiplies across tasks. But then, it follows that outcomes are either lognormal (exp(gaussian)) or Pareto (exp(exponential)). 5/

Item #1 takes many form. Do 10x programmers exist? This is the matter of religious wars, but basically yes, big time. And quant researchers. And entrepreneurs. Here is a good summary: 80000hours.org/2021/05/how-mu… 3/

The two take-aways are: 1. Talented people employed in creative jobs, taken in isolation, have heavy-tailed payoffs. 2. Talented people working together enjoy strong positive externalities, and negative externalities when paired with a few non-talented people. Let's review. 2/

Quand j'étais enfant, on m'expliquait que l'URSS et l'Allemagne de l'Est étaient totalitaires parce que l'État espionnait tous les messages privés. Il y avait des films sur la Stasi, sur l'absence de vie privée et d'état de droit. #chatcontrol

RT @bernhardsson: 100%. I want to short every startup which is trying to be “x for Europe”. Just build a better x!