HNHacker News
TopNewBestAskShowJobs

swyx

24,625 karma · joined December 9, 2015

Writer/Angel in devtools and smol AI enjoyer.

ai engineer pod/news: https://latent.space/

conference: https://ai.engineer

devrel/devtools: https://dx.tips/

personal blog: https://swyx.io/ideas

book: https://learninpublic.org/

but you can also email username [at] ai dot engineer

---

meet.hn/city/37.7792588,-122.4193286/San-Francisco

Socials:

- x.com/swyx - youtube.com/@swyxtv

submissionscomments
swyx··on How LLMs work
agree reasoning fixed a lot of it. but the inherent path dependence of autoregression is one reason i was excited for text diffusion models (https://www.youtube.com/watch?v=r305-aQTaU0)

instead of going left to right, even with a scratchpad, maybe you start with a rough shape of the big picture all at once, and then you iteratively resolve and things come into focus.

mercury (https://www.youtube.com/watch?v=2fDBeMu6xjk) seems to have made the most progress here, which is not saying a ton but is not nothing. i do think it is telling that of the big labs, only GDM has made any meaningful bet on text diffusion. you can bet your ass all of them have evaluated it for a source of alpha.

swyx··on How LLMs work
> If you can see that these models empirically get better with scale, why would you swap the main architecture? Those events will be pretty rare

c.f. hardware lotter https://arxiv.org/abs/2009.06489

swyx··on How LLMs work
not exactly, bitter lesson is one meta-level up from "scale eats everything". this is a common misunderstanding of bitter lesson that rich sutton has been fighting ever since the thing was written. in rich's own words[1], the modern summary is

> Don’t be distracted by human knowledge, as AI has been historically.

> Instead focus on methods for creating knowledge that scale with computation, like search and learning.

so the lesson is choose methods that scale with computation, not just that blindly scaling up anything (data, params, people, whatever) works, it is choosing the right x axis and the right scaling laws consistently wins out in the long run despite short term wins from other methods.

1: https://x.com/RichardSSutton/status/2056419165502935198

swyx··on Show HN: Soft Body Jiggle Physics
why would a "live cam show" company need jiggle physics?
swyx··on Three of our worst VC stories
i mean right but even you hold back some stuff because youre a Nice Person yknow? haha
swyx··on Three of our worst VC stories
theres always a story behind the story. wondering why he chose today to say these
swyx··on Lee Kuan Yew's Singapore Story (2023)
as a singaporean its actually kinda funny to hear everyone here describe him as a neoliberal. not quite our lived experience
swyx··on Gmail thinks I'm stupid, so I left
let me try a different tack - 99% of the value is in the affirmation of the response and the demnonstration of responsiveness, not the actual content value of the 5 words. you could send a thumbs up emoji, you could send a ".", just the "seen" and "acked" value is the value of that 5 word button
swyx··on Show HN: Paseo – Beautiful open-source coding agent interface
good luck sir. if you ever want a job i'm happy to refer you to cognition/devin.
swyx··on Gmail thinks I'm stupid, so I left
then you dont deal in high enough email volume
swyx··on Show HN: Paseo – Beautiful open-source coding agent interface
very nice and well executed product for what i assume is a small team! (of one?)

i guess the obvious question is... if its FOSS and local only, whats the business game plan? you've impressed me that you can ship, but you've basically just matched what eg devin desktop and cursor now do. what next?

swyx··on Chipotlai Max
what i'm hearing is i need to watch this show, it might be the new Silicon Valley
swyx··on Alphabet announces $80B equity capital raise to expand AI infra and compute
google stock is $376 a share rn. berkshire got a favorable discount here. is this... normal? they didnt have an obligation to offer it to the market to find the best price?
swyx··on Chuwi Minibook X
i like how Dave2D came from a nobody to the top cited reviewer here
swyx··on The $500K AI Film That "Premiered at Cannes" Was Not in the Official Festival
i dislike your dismissal, i thought it was well written and researched, and did not equivocate but still gave higgsfield a quote in their own words. if you cant get past superficial writing pattern matches to see that this is better than claudeslop its your issue, not the author's.
swyx··on Claude Code – Everything You Can Configure That the Docs Don't Tell You
also colah was invited by the literal pope to his encyclical https://www.anthropic.com/news/chris-olah-pope-leo-encyclica...
swyx··on Claude Opus 4.8
also useful to have a "chinese wall" between research that knows what went into the models vs marketing/eval models as a third party would
swyx··on Dropbox CEO Drew Houston to step down
hey drew! fwiw it was actually NOT criticism of dropbox specifically - ive worked at and also run a company where this is the case - but for our smaller scale it is because we have support overwhelm and havent sufficiently figured out our own processes to triage everything well. not casting a stone here. you definitely have way more bad actor problems that mean you cant have nice things.
swyx··on Cloudflare Flagship
i see @btown's comment below but also just for education about this space:

- anyone have comments/comparisons about launchdarkly vs posthog vs statsig (is it still alive after openai?) vs _____ vs cloudflare flagship?

like a "beginner/intermediate/advanced" progression of what to look out for/what you will want when it comes to feature flags would be highly helpful for me and many others here

swyx··on Cloudflare Flagship
only 2 hard problems in computer science...
swyx··on Dropbox CEO Drew Houston to step down
sad that going viral on social media is the only functional support system for many tech companies. good luck hope you get your mom's stuff.
swyx··on A sleep-like consolidation mechanism for LLMs
related preprint from the letta team https://arxiv.org/abs/2504.13171

Scaling test-time compute has emerged as a key ingredient for enabling large language models (LLMs) to solve difficult problems, but comes with high latency and inference cost. We introduce sleep-time compute, which allows models to "think" offline about contexts before queries are presented: by anticipating what queries users might ask and pre-computing useful quantities, we can significantly reduce the compute requirements at test-time. To demonstrate the efficacy of our method, we create modified versions of two reasoning tasks - Stateful GSM-Symbolic and Stateful AIME. We find that sleep-time compute can reduce the amount of test-time compute needed to achieve the same accuracy by ~ 5x on Stateful GSM-Symbolic and Stateful AIME and that by scaling sleep-time compute we can further increase accuracy by up to 13% on Stateful GSM-Symbolic and 18% on Stateful AIME. Furthermore, we introduce Multi-Query GSM-Symbolic, which extends GSM-Symbolic by including multiple related queries per context. By amortizing sleep-time compute across related queries about the same context using Multi-Query GSM-Symbolic, we can decrease the average cost per query by 2.5x. We then conduct additional analysis to understand when sleep-time compute is most effective, finding the predictability of the user query to be well correlated with the efficacy of sleep-time compute. Finally, we conduct a case-study of applying sleep-time compute to a realistic agentic SWE task.

swyx··on GitHub Actions down again today
> I haven’t started work yet.

spooky action at a distance

swyx··on Show HN: Git-based front-end interface for Hugo
i did a bit of work on netlifycms, and am a bit disappointed that decapcms's website doesnt seem to acknowledge that history.
swyx··on Show HN: Git-based front-end interface for Hugo
> fills a niche

theres a lot of git based CMSes out there. what is the niche you perceive? (not being critical just want to see what you see)

swyx··on Incident Report: Railway Blocked by Google Cloud (Resolved)
as with many things, we say we like decentralization but quietly vote for centralization
swyx··on Incident Report: Railway Blocked by Google Cloud (Resolved)
i wonder if DID or World (various ways of Proof of Human) can help solve this issue.
swyx··on Incident Report: Railway Blocked by Google Cloud [resolved]
i mean even google and aws are not without sin on this one. maybe wait for an RCA before punching someone who is currently down. theres a reason classy people do "hugops" when a competitor goes down, regardless of reputation.
swyx··on I’ve joined Anthropic
how does this work exactly? do i use that string for other links or how do i change it?
swyx··on Cursor Cloud Agents Down
why is this a HN worthy post?
← PreviousPage 6 of 34Next →