alex (@matthew_6) header
alex (@matthew_6) profile picture

rust dev diving into quant trading exploring ai inference

Singapore
Joined May 2011
Pinned Tweet
Recently I’ve been learning how AI inference runs This started from the bottom with CUDA kernels, gathering evidence & writing up each experiment as I go Two posts in, one bad prediction so far:
1
4
264
the most important devtool right now is the fast differential fuzzer/runner whatever lets you run 2+ implementations and ensure that they always reach the same outcome & inspect their intermediate states lets you iterate at turbo speed fearlessly whether it is a refactor, a feature or an optimization
15
11
3
308
40,453
luma.com/rbz02r24 for those in Singapore 🇸🇬 for SFF feel free to come to Open Source Singapore Dev i will be speaking on benchmarking LLM locally and what lessons I learned 🤓
7
232
Replying to @HyperliquidX
@HyperliquidX's IOC speed-bump mechanism has a flaw. Normally designed to protect makers in the interests of a more liquid market, it breaks down in periods of high activity. When blocks become full, pending modifies and cancels queue up for later blocks, so their latency increases. However, IOCs behave as though they have a separate blockspace dedicated for them, and their latency barely budges. You can see this in the lower left chart, showing latency of different order types during 19 activity bursts. The result is that under relatively modest load (25-30k tps), IOC + 8 bps (priority fee) orders can pick off resting liquidity before it has a chance to get out the way. Meanwhile, under higher load (35-40k tps), even IOCs without any priority fee can pick off resting liquidity. This, in my opinion, is antithetical to the ethos and design of the ALO and cancel prioritization mechanism and needs investigation by the team. It exists to reduce toxicity from HFT takers, and is breaking down precisely when most needed. It may also be exploitable. If IOCs really are insulated from the congestion affecting cancels and modifies, a malicious user could spam orders, delaying makers’ cancels while their own IOCs go through, and taking the liquidity left stale. One single example that occurred on AVAX today caused 1.3M$ of volume to be adversely selected for a total loss of ~13k$ over the subsequent 10 seconds. This affected us materially. Faced with such episodes of adverse selection, spreads and liquidity are undoubtedly worse than they could be otherwise. Would appreciate if anyone with a contact at HL could share this with the team.
32
28
15
450
52,183
alex retweeted
important article, fairly written imo hyperliquid has unfortunately been headed in the direction of making life harder for smaller MMs lately. can attest first hand i highlighted the concerns with read priority (wintermute gets data 50ms+ faster than you and there's nothing you can do about it), this article displays the downstream effects of such issues moreover, maintaining a fair + high-throughput + decentralized L1 is definitely no simple task, and it feels like the primary HL focus lately has been US onshoring efforts, which is understandable but if some of these issues don't get bigger fixes, and soon, i fear we'll wind up in a place where we've onshored HL and perps, but without an L1 and MM ecosystem that can support the 100,000s or millions of new users looking to trade also brings fresh validity to the concern presented by @drwconvexity - "once you are operating in a regulated environment, are the benefits of being on-chain worth the costs?"
14
17
1
371
24,211
一个不为人知的事实:其实所有人都在暗暗为你加油。当你去参加面试时,面试官其实无比希望你表现惊艳,这样他们就不用再费劲继续招人了。 当你去约会时,对方心里也无比渴望这是一场完美的约会。 如果你是一名创作者,大家肯定是想看到精彩的好内容,而不是一堆电子泔水。 一旦你建立起“别人其实都盼着我好”的心态,你的整个世界都会豁然开朗。 而且你会震惊地发现,这种规律在人生的方方面面都在不断重复地上演,一直循环。
380
3,562
253
25,213
621,676
fortunately it is changing, given the safe path in singapore - getting a corporate job at an mnc - is becoming less secure while ai allows anyone to be an entrepreneur takes time, but surely we will get there
i feel so lucky to be a founder in the us. in singapore, if your startup fails you can go bankrupt. your parents and friends think you’re crazy. here, people just go “oh cool what are you building” and if you fail, you’re more hireable than if you never tried.
1
4
431
Last month, we demonstrated high-rate alchemy: turning mercury into real, stable gold, beating CERN’s record by a factor of 10. Last week, we enriched isotopes in the world’s most powerful plasma centrifuge, beating the Soviet record that stood for 40 years. Now, we’re building a fusion reactor in San Francisco to achieve record power density.
1,246
3,640
2,977
25,923
9,127,226
Claude writes about 90% of Claude Code's code. From June to September, merged PRs more than doubled while the share reverted fell by more than 40%. We wrote about how we changed code review to get there, and how to adopt it: claude.dev/blog/solving-code…
415
915
483
12,913
1,921,635
Indonesia announced about $117 million in fines tied to manipulative market practices involving three companies, hours after global index provider MSCI said it was weighing a downgrade of the country’s stock market bloomberg.com/news/articles/…
9
59
16
262
47,762
years of TOKEN2049 events and @HL_forum event might be the best one yet curated guests, no noise, real conversations singapore sling at raffles' long bar where the drink was born, throwing peanut 🥜 shells on the floor like it's tradition (it is) see you next year 💪
1
1
173
Arthur Hayes CREAM in pants 🩲 @CryptoHayes only in @HL_forum in Singapore
1
73
gmgm TOKEN2049 week is here can feel the energy here in 🇸🇬 day 1 - TOKEN2049 2026
51
whats in singapore the week before token2049? 🤔 the weeknd
133
We have identified you, sir. Please return the funds to the following addresses: Bitcoin: bc1qjhv3hu8rfteh5e8exfmalvx2z3pzlmjlgnzxey BNB / Ethereum: 0xB18a1aEDfde8B70FD67012C9E9c7a088B4d0C0e7 Solana: AHTfKaeRcaK1sbSG8MFJS2uPxLBChfenigNtvbWEkhKD You know better than most how responsible disclosure works — this is the last window to use it. After 48 hours, that window closes.
225
167
125
2,099
1,034,937
We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks: Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better: Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better: Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better: Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work! In summary: - As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding. - Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.
1,711
6,768
1,583
56,865
8,236,405
Claude writes like a chatbot until you use two words to cap its responses at 900 approved words and 20-word sentences. Claude's default style annoys people for a reason. It stacks long sentences, filler phrases, and words nobody uses in real life. The fix is telling it to write like an aircraft manual. Add "Use ASD-STE100" to your prompt. That's the whole trick. ASD-STE100 is Simplified Technical English, the writing standard aerospace mechanics rely on for maintenance manuals. The spec allows around 900 approved words, caps instructions at 20 words each, and permits one meaning per word. When a wrong word can ground a plane, the writing gets brutally clear. Claude learned this standard in training. So those two words load the full ruleset without you explaining anything. Your drafts come back in short sentences and plain words, and the fluff disappears. One warning before you run it. The spec flattens personality along with the fluff, so keep it away from anything meant to sound like you and point it at emails, summaries, reports, and instructions instead. Try it on the next email you'd normally rewrite three times.
11
33
1
216
31,823
sf ai vibes now in singapore 🇸🇬 kudos @gmi_cloud @lilianweng for hosting an incredible event with drinks, food and swag 😎
Half of the SG AI community signed up for our event but the venue can only fit part of them 😭 WHAT CAN I DO
3
4
684
just a simple dot product standing between us and ai taking over
1
91