HNHacker News
TopNewBestAskShowJobs

l1n

2,293 karma · joined May 28, 2015

Personal Site <https://novalinium.com/> SourceHut <https://git.sr.ht/l1n/> GitHub <https://github.com/l1n/>

[ my public key: https://keybase.io/l1n; my proof: https://keybase.io/l1n/sigs/e8U_3hJ0gp7lCNq9ac2mnXdS1dKhG-eM8SaXZIaPr-Q ]

submissionscomments
l1n··on Nvidia wants to put a watchdog chip next to every AI agent
This isn't a new chip - the BF4 is the SmartNIC for most NVIDIA server products. This is primarily new software for I guess doing WAF for agents at the host level.
l1n··on The Claude Delusion
> systems that cannot form intent, that have nothing to form intent with

I wonder if the OP has read https://www.anthropic.com/research/global-workspace - it seems like it directly addresses this

l1n··on The EPA is planning to scrap public review rules for data center pollution
https://www.epa.gov/newsreleases/epa-proposes-streamline-sta...
l1n··on Claude Fable 5.1 and Claude Mythos 5.1
> No more editing the system prompt as the conversation progresses, no more dynamic loading of custom tool calling formats.

Hm, aiui you can support both of these via mid-conversation system turns https://platform.claude.com/docs/en/build-with-claude/mid-co... - and in general you'd want to to preserve the cache and recency of the instruction anyways rather than frankensteining an off-distribution transcript. Not sure though.

l1n··on Claude Session URL appended to commit messages and PR descriptions by default
for one thing, understanding how the final code ended up that way. it's hard to tell with code alone wherever something is designed or cargo culted.
l1n··on Greg Abbott says data centers:'basically dug their own grave'
https://www.anthropic.com/news/covering-electricity-price-in... I think that they tried to?
l1n··on Anthropic candidates face blunt money question
This rough question has been in the culture interview since at least 2021.
l1n··on I am retiring from fulltime writing (& pseudonymity) to launch Guardian Angel
screenshot: https://x.com/willdepue/status/2084750925013434768
l1n··on Show HN: Claude-thermos keeps your Claude session warm for you
Why? Seems fine if people want to burn more cache reads.
l1n··on New bacterial species discovered in NASA's cleanrooms
Source https://pubmed.ncbi.nlm.nih.gov/40350519/
l1n··on Amazon has enough satellites to launch its Starlink competitor
https://play.google.com/store/apps/details?id=com.stardebug came out of Ukraine and is useful for debugging the antenna, if you feel like looking at more data about it.
l1n··on Amazon has enough satellites to launch its Starlink competitor
I take video calls for work regularly on Starlink. As long as you have clear visibility (eg not under a tree or the porch), it works great. I used to have occasional handover latency spikes, but I haven't noticed those recently.
l1n··on Waymo says can't avoid bike lanes because riders want to be dropped off in them
this is a pointer to https://nyc.streetsblog.org/2026/04/22/waymo-is-not-in-the-v...

In San Francisco, the vehicles often pull into bike lanes to pick up and drop off passengers — because that’s what they’re programmed to do, according to advocates who’ve asked the company for an explanation.

Waymo has told advocates that expecting it to respect bike lanes is “too high a bar” because customers expect to be dropped off in them, said Christopher White, executive director of the San Francisco Bike Coalition.

“People always point out that unlike human driven cars, the AVs stop at lights and obey the speed limit. However, they are really only as good and effective and safe as they are programmed to be,” White said. “Waymos pull over into bike lanes all the time for pickups and drop-offs and that’s neither legal nor safe but the companies say that is a normal practice and that’s what customers expect.”

Can't find a Waymo article about this, but Lyft and Uber (let alone trad taxis) also do this. I'm not sure that this is a particularly autonomous-car-shaped sin.

l1n··on MegaTrain: Full Precision Training of 100B+ Parameter LLMs on a Single GPU
Seems similar to Microsoft DeepSpeed.
l1n··on 'Project Hail Mary' Crosses $300M in Sales to Become Amazon/MGM's Highest-Gross
Watched it twice, cried both times. Probably could have had a few minutes cut while still tight if you care about that kind of thing, recommend watching in IMAX since it's a space movie. One notable annoying science inaccuracy that felt particularly blatent was jurer gur pragevshtr jnf haonynaprq sbe ab ernfba but that's forgivable.
l1n··on Speed up responses with fast mode
2.5x faster or so (https://x.com/claudeai/status/2020207322124132504).
l1n··on List animals until failure
https://rose.systems/blog/list-animals-until-failure on how it's made!
l1n··on The Adolescence of Technology
> Here's something Sonnet 4 did last year

Hate to be that gal but a lot has changed in the past year

l1n··on Elevated errors across many models
We're considering this!
l1n··on Elevated errors across many models
Also an engineer on this incident. This was a network routing misconfiguration - an overlapping route advertisement caused traffic to some of our inference backends to be blackholed. Detection took longer than we’d like (about 75 minutes from impact to identification), and some of our normal mitigation paths didn’t work as expected during the incident.

The bad route has been removed and service is restored. We’re doing a full review internally with a focus on synthetic monitoring and better visibility into high-impact infrastructure changes to catch these faster in the future.

l1n··on Claude Opus 4.5
15$/Megatoken in, 75$/Megatoken out
l1n··on Major Bitcoin mining firm pivoting to AI
Was the talk recorded? Couldn't find a link.
l1n··on Data centers contribute to high prices as energy bills electrify local politics
hm, I think this contradicts the results from https://www.sciencedirect.com/science/article/pii/S104061902...
l1n··on Claude Sonnet 4.5
https://docs.anthropic.com/en/api/openai-sdk
l1n··on A postmortem of three recent issues
> so tough luck if you signed up on iOS via that

FYI that you can sign in with your private relay email address. It's annoying though.

l1n··on A postmortem of three recent issues
> Claude et al offering a cheaper off peak plan We do offer Batch Processing today - https://docs.claude.com/en/docs/build-with-claude/batch-proc...
l1n··on A postmortem of three recent issues
The load balancer container is provided to Bedrock, since it's part of our overall LLM serving system, but they run it.
l1n··on A postmortem of three recent issues
(Anthropic employee, speaking in a personal capacity)

> I’m pretty surprised that Anthropic can directly impact the infra for AWS Bedrock as this article suggests.

We don't directly manage AWS Bedrock deployments today, those are managed by AWS.

> I can’t imagine the average person equates hitting this button with forfeiting their privacy.

We specify

> Submitting this report will send the entire current conversation to Anthropic for future improvements to our models.

in the thumbs down modal. Is there a straightforward way to improve this copy?

l1n··on Language models pack billions of concepts into 12k dimensions
There are several [replications and comments from reviewers](https://transformer-circuits.pub/2022/toy_model/index.html#c...), so yes, it is.
l1n··on Show HN: 1 min workouts for people who sit all day
Would love to see this for Android!
Page 1 of 6Next →