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 ClickHouse acquires Langfuse
(congrats team! such a joy to see you succeed)

every single day there is an acquisition on here. what's going on in the macro?

swyx··on CVEs affecting the Svelte ecosystem
all DoS attacks and one XSS. this isnt as bad as the react server components CVEs, which enabled RCE.

saving people a click:

CVE-2026-22775: DoS in devalue.parse due to memory/CPU exhaustion

> Effects: A malicious payload can cause arbitrarily large memory allocation, potentially crashing the process. SvelteKit applications using remote functions are vulnerable, as the parameters are run through devalue.parse If you don’t have remote functions enabled, SvelteKit is not vulnerable

CVE-2026-22774: DoS in devalue.parse due to memory exhaustion (Yes, this is very similar to the previous CVE. No, it is not the same!)

> Effects: A malicious payload can cause arbitrarily large memory allocation, potentially crashing the process SvelteKit applications using remote functions are vulnerable, as the parameters are run through devalue.parse If you don’t have remote functions enabled, SvelteKit is not vulnerable

CVE-2026-22803: Memory amplification DoS in Remote Functions binary form deserializer

> Effects: Users can submit a malicious request that causes your application to hang and allocate arbitrarily-large amounts of memory

CVE-2025-67647: Denial of service and possible SSRF when using prerendering

> Effects: DoS causes the server process to die SSRF allows access to internal resources that can be reached without authentication from SvelteKit’s server runtime If the stars align, it’s possible to obtain SXSS via cache poisoning by forcing a potential CDN to cache an XSS returned by the attacker’s server (the latter being able to specify the cache-control of their choice)

CVE-2025-15265: XSS via hydratable

> Effects: Your users are vulnerable to XSS if an attacker can manage to get a controlled key into hydratable that is then returned to another user

swyx··on I hate GitHub Actions with passion
fyi i maintain a repo that accidentally tracks github actions cron reliability (https://www.swyx.io/github-scraping) - just runs a small script every hour.

i just checked and in 2025 there was at least 2 outages a month every month https://x.com/swyx/status/2011463717683118449?s=20 . not quite 3 nines.

swyx··on There's a ridiculous amount of tech in a disposable vape
as a first step, let's tax these things. this is such an immense waste of electronics.
swyx··on Exa-d: How to store the web in S3
hi will! super nicely written, nice look under the hood of your processing. as an orchestration guy i always wondered why everyone seems to converge on using Ray, and as a secondary thought, how well is Anyscale capturing the Ray market.

if i were doing what you do i might set up a lot of rate limits/anomaly detection in case some weird unintended invalidation causes a weird spike in your dependency graphs. is there good practice there for anomaly detection other than "setup a bunhc of dashboards and be on call"?

swyx··on Sampling at negative temperature
> . I've experimented with this as a metric for distinguishing whether text is written by AI. Human-written text doesn't follow a constant-temperature softmax distribution, either.

oo that sounds like a cool insight. like just do a trailing 20-30 token average of estimated temperature and look for variance like one might do a VO2 max

swyx··on Sampling at negative temperature
"What might be more surprising is that even when we adjust the temperature down to 0This means that the LLM always chooses the highest probability token, which is called greedy sampling. (thus making the sampling theoretically deterministic), LLM APIs are still not deterministic in practice (see past discussions here, here, or here)"

https://thinkingmachines.ai/blog/defeating-nondeterminism-in...

swyx··on Sampling at negative temperature
interesting exercise and well written. my followon questions/work would be:

1a. temperature=100000 is interesting too. obviously "ideal" temperature lies somewhere between 0 and 100000. has anyone ablated temperature vs intelligence? surely i'm not the first person to this idea. commonly people try to set temp=0 to get "deterministic" or "most factual" output but we all know that is just Skinner pigeon pecking.

1b. can we use "avg temperature" as a measure in the way that we use perplexity as a measure? if we see temperature as inverted perplexity with some randomness thrown in, are they basically the same thing inverted? or subtly different?

1c. what's the "avg temperature" of most human communication? whats the "avg temperature" of a subset of "good writers"? whats the "avg temperature" of a subset of "smart writers"?

2a. rerun this negative exercise with constrained vocab to english

2b. RL a model to dynamically adjust its own temperature when it is feeling 1) less confident 2) in brainstorm mode

2c. dynamically inject negative temperature every X tokens in a decode, then judge/verify the outcome, to create high variance synthetic data?

its hard for me to follow the train of thought on 2 because negative temp is essentially not that different from ultrahigh temp in practice.

swyx··on How Markdown took over the world
sharing my list of mistakes in markdown that Gruber endorsed :) https://news.ycombinator.com/item?id=22776108
swyx··on Bose has released API docs and opened the API for its EoL SoundTouch speakers
no, they were not made to do it. they listened to feedback and did the work. this is better than we get in 99% of cases. try to be nicer and meet them half way instead of living in your ideal world.
swyx··on Show HN: Similarity = cosine(your_GitHub_stars, Karpathy) Client-side
the frontend is beautiful. i find it inspiring that you have 10 years of data science and are no longer limited by your lack of frontend or design knowledge. this is a better site than i couldve done
swyx··on The Jeff Dean Facts
> Jeff & Sanjay did everything together,

my nonexpert impression is jeff keeps much more of a public profile. hence the natural celebrity goes to him. was this not true way back in the day?

swyx··on LMArena is a cancer on AI
i asked them this in my interview. tldr they subsidize all inference on their platform https://www.youtube.com/watch?v=NBnOk0Uy9ig&t=70s
swyx··on 65% of Hacker News posts have negative sentiment, and they outperform
> The results I use in my dashboard are from DistilBERT because it runs efficiently in my Cloudflare-based pipeline.

interesting - why use cloudflare vs say hf inference or modal? and is this replicate-cloudflare or normal cloudflare?

swyx··on The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks (2018)
he pretty much always says it offline haha but i maay have mixed it up with the subsequent convo we had at neurips https://www.latent.space/p/neurips-2023-startups
swyx··on The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks (2018)
i intereviewed Jon (lead author on this paper) and yeah he pretty much disowns it now https://www.latent.space/p/mosaic-mpt-7b
swyx··on Databases in 2025: A Year in Review
look up the cmu db youtube
swyx··on Databases in 2025: A Year in Review
very interesting. whats the vector indexing story like in duckdb these days?

also are there sqlite-duckdb sync engines or is that an oxymoron

swyx··on Lessons from 14 years at Google
> 17. Your network outlasts every job you’ll ever have.

i mean, addy has literally been at google 14 years, i think internal network has outweighed the external one here

swyx··on Claude Code On-the-Go
yeah this was a poorly written blogpost (which is fine to be clear, something is better than nothing and OP gave the highest order bits).

but its this poke: https://poke.com/ verified because TFA is cited in this page https://poke.com/explore

the sign up is very annoying fair warning.

swyx··on Eurostar AI vulnerability: When a chatbot goes off the rails
slander is again a different thing. you should also be more careful using those loaded words.
swyx··on ManusAI Joins Meta
have had 2 outsider estimates (1 public, 1 not, both more well informed than avg HNer) that acquisition was ~$4B worth, def not play money.

i just released the full AIE workshop covering Manus' product surface area if anyone is also out of the loop and wants to catch up: https://www.youtube.com/watch?v=xz0-brt56L8

(no vested interest am just friends w Ivan who works there. also as a singaporean i guess this is a small W for the Singapore AI scene)

swyx··on Show HN: HN Wrapped 2025 - an LLM reviews your year on HN
you should probably just make this your bio
swyx··on CEO killed at industrial site by worker operating forklift
ok but also something is still not adding up here - sure the operator was distracted, but you a presumably functional CEO are crossing the road, and you cant hear a forklift moving/dont think to look like at all? these things dont move that fast esp on a worksite
swyx··on Ask HN: Why isn't there competition to LinkedIn yet?
basically instead of treating linkedin like a chore to minmax, you are genuinely trying to reach out to 1) people you are trying to hire, 2) buyers/decisionmakers who have the pain points you solve. linkedin is very good for that.
swyx··on 10 years bootstrapped: €6.5M revenue with a team of 13
very nice! i like seeing Heroku and Digital Ocean hang in there and win
swyx··on Local AI is driving the biggest change in laptops in decades
> state-of-the-art models

> hundreds of millions of parameters

lol

lmao, even

swyx··on 10 years bootstrapped: €6.5M revenue with a team of 13
> We spent the winter prototyping (experimenting with everything from bare metal to alternative PaaS providers — some of which shall remain unnamed to protect the guilty

mind at least naming what to look out for when evaluating PaaSes?

swyx··on 10 years bootstrapped: €6.5M revenue with a team of 13
> we achieved an EBIT margin of 65%

isn't it sensitive to disclose this kind of info when you don't have to? are you worried about your employees all demanding raises when they see this?

swyx··on GLM-4.7: Advancing the Coding Capability
> Preserved Thinking: In coding agent scenarios, GLM-4.7 automatically retains all thinking blocks across multi-turn conversations, reusing the existing reasoning instead of re-deriving from scratch. This reduces information loss and inconsistencies, and is well-suited for long-horizon, complex tasks.

does it NOT already do this? i dont see the difference. the image doesnt show any before/after so i dont see any difference

← PreviousPage 13 of 34Next →