Hacker News Reader

persedeson Just a little AI resistance, as a treat

it is very good, but keep in mind it'll also flag text edited with grammarly. Maybe not the same as copy and dumping a claude response, but still normalizes the authors voice with AI.

persedeson BYD sold more EVs last month than the US did all quarter

Hah was wondering if someone else had read that announcement/ finding. Coupled with this https://www.theatlantic.com/international/2026/10/chinese-co... (lots of dead companies being propped up artificially by the govt and not allowed to fail so people stay employed) the strategy of cornering every global export market until complete dominance seems to be very a much make it or break it approach here. So the US blocking one important industry with likely lots of subsidies does make more sense under that light.

persedeson Anthropic reports $17B in Q3 revenue, commitments of $895B in revised S-1

Yeah I think for the last 3 months or so, the only ones that are still token maxing are the ones on subscriptions.

persedeson Herdr has released native app

True, but tmux is very configurable

https://github.com/hiroppy/tmux-agent-sidebar

Has been pretty nice

persedeson Herdr has released native app

I was looking into Herdr too, but could not bring myself to switch away from tmux yet (there are a couple of shortcuts that I have in my workflow, which I can't config in Herdr). What I ended up doing was to use the tmux agent sidebar, which, with mouse support, I think does a lot of the same things already.

persedeson Binance-hacker-1029 – $5.6B stolen from Binance

I'm not in the space, but I'd think that the reputational damage alone is the more significant part here.

persedeson Dayjob: Tax – evaluating AI agents on 50 professional tax assignments

There's also: https://github.com/column-tax/tax-calc-bench

persedeson Dayjob: Tax – evaluating AI agents on 50 professional tax assignments

https://claude.com/marketplace/connectors/aiwyn-tax

Try this, it uses a tax engine as the backbone and provides feedback for Claude if it goes off the rails. Albeit it doesn't support every tax scenario.

persedeson Wikipedia's Cleanup of AI-Generated Content

They're providing the training data for free too. I'd think there's mutual interest there.

persedeson Wikipedia's Cleanup of AI-Generated Content

100%. Looks like they're using pangram (hope they're getting that for free...) which also flags stuff like Grammarly. That one's an interesting gray zone to me that I've not thought off before. LLMs whales doing the heavy lifting is one thing, but small edit tools like grammarly that "improve" your phrasing are likely a massive algea like contributor to what is out there too.

persedeson eMudhra to exit global SSL/TLS certificate business

It's 0x0000000000000000000....4, but no, just good ol' human failure (and a bunch of extra f-ups along the way)

persedeson Alternatives to animal testing are finally taking off

Nice to see TissUse getting a shout out :)

  the FDA wanted data from organs-on-a-chip or some comparable alternative.

This is the main takeaway from the article imho and great news! ~15 years ago those systems were were not very interesting for pharma companies (bayer used to have a campus a few s-bahn stops away) due to regulators requiring animal tests over those systems.

persedeson If you don't run your own LLM proxy, how do you know whats going on?

I think businesses are footing the majority of that bill right now due to regulatory requirements, but for individuals that price increase is def coming. Maybe I'm too pessimistic, but enjoy it while it lasts :')

persedeson I hired an illustrator to draw my house. Now it's my Home Assistant dashboard

Hah second the AI part. Started a home assistant setup years ago to setup water leak sensors with alerts to my phone, but didn't feel comfortable enough having home assistant accessible from the web, so besides writing brittle yaml files now also had to deal with open VPN etc etc and after major update my config files were incompatible and with kids just didn't get back to it.

However this week spent an hour with claude to migrate to tailscale, latest home assistant update, sensor setup and filtering for my rtl433 antenna, connected hue lights while we're at it, migrated to latest pihole version and put everything in git in one nice compose file with a step by step instruction on how everything was setup and which account sign ups were needed. I still checked every step and manually approved stuff, but the remote session feature made this easy to finish.

persedeson OpenAI withdraws three mathematical results

  > All the human needs to verify is that the statement of the theorem is translated correctly from natural language to Lean. That usually covers a very small surface of the Lean code.

That's exactly what the navier stokes paper posted yesterday pointed out where the LLM bends the Lean code to make it "compile", because the NL might be wrong to begin with or because it missed a detail:

https://arxiv.org/html/2610.08144v1#S2

For complex / tedious proofs I can easily see how small details like this can lead to a valid lean proof (or valid "code"), but missing the important details that got lost.

persedeson Navier–Stokes Lost in Translation

Trying to map this observation to code and appreciate them starting with very simple examples (that Astra screw-fixed into lean.) But in a simplified way this is close to promting the model to create a set with the members 1,2,1,5 and it correctly creates {1,2,5} in code. So The initial "proof" / instruction was wrong and it silently fixed that. ( Which is one of the failure modes they're describing)

persedeson Claude Haiku 5.5

Or just for the token usage. The rug pull is coming eventually

persedeson Improper redaction reveals Google Data Center water and electricity usage

That's wishful thinking, I know, but would love a breakdown of those metrics into what actually goes into AI inference, training versus just TikTok / YouTube streaming, or just a Google search...To have some perspective. My hypothesis that prolonged video streaming might contribute similarly to AI usage? (But CDN etc might make that tricky to calculate...)

persedeson Sonnet 5.5

Those models are cheaper per token, but depending on your use case you might still end up paying more with the cheaper model. DeepSeek likes to burn through a lot of tokens for example, which can quickly ameliorate those savings. They are great models and have most likely helped anthropic and openai drop their prices lately, but I still don't see the monetary benefit of using them atm.

persedeson Stripe's Knowledge AI Platform

my biggest code smell is debugging. Trying to step through the code is slow because of how much the debugger has to keep track off and it's a lot of step into / jumps until you get to the piece you actually care about. Then you also see stuff like "if langchain_v2: <branch> elif langchain_v3: <branch> else ...." ... breaking changes are ok! I feel like at this point they're hopelessly spaghettied and need a more opinionated approach to clean up a lot of the mess.

Since others are asking: My current goto is pydantic ai and it's been doing pretty well for my use cases.

persedeson Stripe's Knowledge AI Platform

Cool to see, but wondering what the upside of this is vs what notion is building with their own agents. I can imagine notion agents can do something very similar to this (and with the same amount of control / security) and at what point would it make sense for a company to build their own vs buying into one..

persedeson Claude Opus 5.5

how are they exaggerating the results? Comparing the cost from that chart for 5 and 5.5 for medium-max effort paints a pretty clear picture:

         mean  median
 model
 5      4.135   4.245
 5.5    3.150   2.640

Again seeing how max is a clear outlier, the median cost saving is ~38%, not that far off from the proclaimed 40%.

persedeson Disney+: New user agreement allows ads before movies in all subscriptions

I try to post this link everytime someone praises libby :|

persedeson Claude Opus 5.5

so don't use it at max? The benchmarks suggest that high/xhigh are more than sufficient to be ahead and a whole magnitude below max with regards to token usage. I'd treat that as an outlier and not how verbose the model is in general (QED I know)

persedeson GPT-6 Sol and Luna

Not that anthropic models are very good at this, but due to the changes in tokenizers and thinking tokens: cost per token is not as helpful anymore as cost / task.

persedeson OpenAI GPT–6 Astra breaks Enigma message that has resisted solution since 2005

Hmm unless there's regional differences, but "shove after leader" would be more accurate. "Hinschub" could be translated as shove towards.

persedeson Disney+: New user agreement allows ads before movies in all subscriptions

For now at least:

https://ninelives.karawynnlong.com/p/the-coming-enshittifica...

persedeson Notes on gotchas while migrating 35kb preprompts from Opus to self-hosted Ollama

RAG was supposed to be the way out on that, and ended up being mostly abandoned.

Unless you're a contracting agency!

persedeson Coyote v. Acme (1990)

Did wonder about the review, it probably sold well in the news papers and part of me thinks that the Odyssey team might even agree with her, but IMHO it's mostly overblown? The movie is good an wildly successful, and maybe I've gotten less jaded, but felt like her criticism did not land ( saying that as someone who loves to poke holes at things and reading a good polemic).