HNHacker News
TopNewBestAskShowJobs

hobofan

8,360 karma · joined March 21, 2014

Jack of all trades (- master of some?)

goisser94@gmail.com

https://hobofan.com

Mastodon: @hobofan@toot.berlin (Twitter: @hobofan)

---

Currently building https://github.com/EratoLab/erato

submissionscomments
hobofan··on On the Navier–Stokes Millennium Prize Problem
The distiction they are trying to make is: "One of our employees or the model was able to verbatim read the chats when they were actively tackling the problem" vs "The chat of someone working on the problem may have ended up in the training set of the model".
hobofan··on LibreOffice breaks download records after declaring it has no AI features
https://simonwillison.net/2026/Sep/1/codex-libreoffice/ - https://news.ycombinator.com/item?id=49527396

(and I can also just add that it's my go-to tool for that exact use-case, and used it in a dozen different client projects)

hobofan··on Mistral raises €3B
Regulation != censorship. (Apart from that, Mistral has Shieldstral[0] for adding policy enforcement).

With it's general approach of Open Weights, and being able to be deployed on-premise (/private/public cloud), they are a viable solution for your typical European enterprise, that has to comply with traditional (non-AI) regulation. With the NIS2 directive the amount of those companies is also significantly expanding.

We[1] are in the same market as Mistral, and among our customers, the go-to-solution of MS Copilot is typically performing badly, and the typical SaaS solutions are not even given a consideration, which is why they are reaching for on-premise-first solutions.

[0]: https://mistral.ai/news/shieldstral/

[1]: https://github.com/EratoLab/erato

hobofan··on Mistral raises €3B
Are they, though? I would say, let's find out when the IPO settles, but then again the market also sustains a incredibly crazy valuation for Tesla.
hobofan··on I'm a seeing-eye dog for a computer
Rerun is pretty dope, but I'm not sure how you came to the conclusion that it's the "author's main project"? There is a whole company backing it, with no affiliation that I could find, apart from the author being an occasional contributor?
hobofan··on I Feel about AI
All individual robotic parts are already far enough in research. Even if they aren't, an AI with enough resources can continue doing that research on its own.

It could likely get a good leg up by breaching the security of all top robotics labs and exfiltrating their documents.

Ultimately, if AI is advanced enough, and it were to decide to compete with humanity, there is essentially nothing that can be done to prevent it from embodying itself. As long as there is a single rack of GPU servers that it can hack into anywhere in the world that is unsupervised enough to where it can escape detection, there is no way to stop it. This would require an unprecedented (unrealistic) level of cooperation of all humanity to achieve.

hobofan··on How I feel about AI
> we have full control over the tool call dispatching

> cannot do anything without the harness doing the execution

This only holds true as long as the harness is exploit-free. A sufficiently advanced AI can in theory (and I think there recently were some POCs showing something like that) break the containment that the harness creates, if e.g. there are vulnerabilities in the tool call parser.

hobofan··on How I feel about AI
> how is an AI going to affect anything in the real world?

Robotics

hobofan··on GPT-6 Astra on robot arms
> People hire cleaners and tradespeople into their homes all the time.

It's very obvious that there is a difference between a single person that you can personally vet, and something that records videos that can be shared anywhere.

> I think internet people care about some theoretical surveillance angle more then anyone else actually does.

Yes, that's usually the case. However in the very current time of surveillance resistance with the de-Flock movement, I think there is increased mainstream awareness of the issue.

hobofan··on Cloud in a Bottle: making self-hosting accessible to everyone
> Only Gen Z and later are starting to care and willing to spend the time and effort to self host.

I'm not sure what you are seeing, but what I'm seeing is widespread teach illiteracy in Gen Z outside of their smartphone. Gen Z and younger mostly grew up with smartphones as their primary devices and with little contact with a laptop or desktop computer (and even then confined 99% to browser usage).

hobofan··on Wikimedia Foundation Workers Overwhelmingly Vote to Form Union with CWA
Unions are not just for bad employers.
hobofan··on Hackers had a live feed of every ID verification company scanned for over a year
The HN title is misleading.

> Hackers Had A Live Feed Of Every ID __This__ Verification Company Scanned. For Over A Year.

The "This" in the the sentence serves an important role. It currently reads like all ID verification companies were compromised at the same time.

hobofan··on Authorization terminology is a mess: Let's fix it
I'm not opposed to AuthS either! :)
hobofan··on Authorization terminology is a mess: Let's fix it
This has already been solved well-enough with AuthN and AuthZ as distinct names.
hobofan··on Why humanoid robots won't catch up to human workers any time soon
Even with just teleoperation, the first robotics companies are poised to have some of the most efficient in-field technician operations, as long as there are at least two robots on site.
hobofan··on Saving money on Google Photos with Immich: Your own personal photo storage
It is possible in general, but almost everybody doing self hosting (and as outlined in the article) does it on a local network and not on the internet. Most people that self host don't want to have to be a sysadmin in their free time in order to have a reasonable risk of not leaking all your families photos to the internet.
hobofan··on Why humanoid robots won't catch up to human workers any time soon
I think the article takes a completely wrong lense for viewing the Humanoid Robotic Games, and general progress that is being made right now (though admittedly, many robot companies make it challenging with their marketing as well). Most of the advances being demonstrated right now are about establishing a robotic _platform_. They don't necessarily individually have to be able to complete all tasks end to end, but pushing the boundary forward on one of them at a time as a step of progress.

The different challenges presented in the article are almost all orthogonal to each other. Some of them are hardware engineering challenges, some of them are software / AI challenges, and most of them ultimately economic challenges. They can all be solved in a mostly independent manner, meaning that they won't be decades away to be resolved, but more likely years.

I think robotics is in a space that very much falls into the "people overestimate the progress in one year but underestimate the progress over 5 years" category.

A lot of the problems to solve regarding effectiveness (not efficiency) of humanoid robots even as outlined in the article (manipulation = short term spatial planning; long term task planning) greatly benefit from the current advances in AI, and in my estimation are already solvable with better harnesses with the current generation of models.

hobofan··on It takes 5 cloud services to hear my doorbell
I presume you want to keep the Blink doorbell for the camera functionality?

Then I guess 3d print a casing that just puts the ZigBee button on top of the integrated button?

hobofan··on Domestic cats kill 1.3–4.0B birds and 6.3–22.3B mammals annually
A prerequisite for there to even be a chance of cats being involved in ecosystem balance would be for the cats to exist purely as part of the ecosystem, as in being exposed to negative feedback effects to it (e.g. starvation due to overhunting). As it stands, they are being fed by humans and mostly exist outside the ecosystem, and every kill by them only drains resources from the ecosystem.
hobofan··on My Business Is Dying
I can only tell you what we see across our customers, which are exactly the on-premise crowd. We are building a ChatGPT/Cowork competitor focused on on-premise needs[0]. While we cover most things outside of the coding side of things, I know that all of them are also embracing coding agents, typically with your usual open weight coding models hosted on-premise.

Sure, the slow behemoths that you talk about exist as well, but I am seeing many companies that have successfully turned compliance into just another kind of process that isn't slowing down the product development lifecycle. And within those process steps they are also able to leverage AI to speed up e.g. review processes.

[0]: https://github.com/EratoLab/erato

hobofan··on My Business Is Dying
Most places I know that have strict on-premise/air-gap requirements are just investing in on-premise AI and open source harnesses instead. The only ones that don't are the typical places that are slow to adopt technology in general (usually because they can afford to due to historic moat).
hobofan··on AWS Acquires DuckLabs
> Google would drive them to quit over a longer timeframe, with no threats or harrassment, just because every time they try to ship something cool, someone else has a reason not to do it and that kills you eventually.

Yeah, I've never personally worked at any of those places, but collaborated on a few open source projects with Googlers and Xooglers and the slow grinding wheels of their "processes" that wear people down over time were very apparent. Nevertheless, it has been a breeding ground for many interesting technologies, even if it often suffocates them over the long term.

hobofan··on AWS Acquires DuckLabs
Datafusion has all of the individual pieces, especially if you just want to data analysis, but is far from an embedded database.

Indexes, transactions, a first-class storage format are all things that come included with DuckDB, that you won't have with Datafusion.

hobofan··on AWS Acquires DuckLabs
Fig, was one of them but was only partially opens source IIRC. CLI autocomplete and script team sync. Turned into/integrated into Amazon Q, which turned into Kiro CLI, which AFAIK doesn't share any functionality with the original product nowadays.
hobofan··on AWS Acquires DuckLabs
> I’m not sure how much people think open source companies are worth, but the path to VC-scale profitability is incredibly rare.

MotherDuck, a semi-competitor of theirs has raised $100mil in funding. DuckLabs has reportedly taken no external funding (so all ownership is with the founders) and was profitable with 30+ employees.

On the sidelines, other companies behind beloved open source products, like Astral, Astro, Bun are getting bought left and right.

With that as a backdrop, I think they should have been able to get quite a good payout.

hobofan··on AWS Acquires DuckLabs
AFAIK Google is much better at not tearing well-working teams apart in reorgs, and letting people go they have acquihired a few quarters earlier.

Of the few examples I have I my head, I'd even say that the fate of a product that has been acquired by Google is probably even better than those of Google's internally developed products. e.g. Waze is still alive and kicking 13 years after acquisition under its own brand and hasn't been completely swallowed by Google Maps. The Nest brand also stuck around for quite some time.

hobofan··on LLMs could control their host machines by exploiting inference engines
I'm not saying that the HTTP interface is not an attack surface, but that securing that attack surface (as one may assume in a simple security model) is not enough to protect against that class of attacks, as no level of introspecting the input will help predict and secure against malicous output.
hobofan··on AWS Acquires DuckLabs
I'm glad that DuckDB has a foundation in place and hope it is resilient enough to push the DB forward when the time comes.

Out of all the big orgs, Amazon is probably the one that has the least regard for keeping technically interesting projects alive, and the certainly will bulldoze it for some dumb reason when the next re-org comes.

hobofan··on Solving the 1+N Query Problem
Yes, this does essentially nothing to solve the 1+N query problem. If your solution was to to stuff everything into one query, this was already possible with SQL!
hobofan··on LLMs could control their host machines by exploiting inference engines
_you_ wouldn't. The blog post is about vulnerabilities that happen inside the inference engine, and not via external interfaces accessible via HTTP. The token parser is involved both on the way in (where yes, you would as user supply it likely via HTTP), as well as the way out of the LLM (where there are very few, less hardened layers when interfacing with the parser).

e.g. for a vulnerability exploit in the parser, an LLM can be trained with a specific input sequence that would output a fixed malicious payload to output eval-able content, as outlined in the article.

← PreviousPage 5 of 34Next →