8,360 karma · joined March 21, 2014
goisser94@gmail.com
https://hobofan.com
Mastodon: @hobofan@toot.berlin (Twitter: @hobofan)
---
Currently building https://github.com/EratoLab/erato
100% up to the harness. Most harnesses either fixed (or dymanically depending on size) nowadays add a "search_tool" tool to prevent spamming the context with all tools.
While it may intuitively may look like there is a large overlap in the three areas there is suprisingly little, and for the few parts there are (e.g. trace <-> log correlation), OTEL does offer a standard.
Maybe I'm biased from the perspective of a "harness provider", but I think Model routers often have too little context to act as well-informed prompt injection prevention. e.g. it lacks context of where which part of the message(s) originates from and sanitization/safeguards were already performed on the application layer.
Something like OpenRouter's "flag" mode is fine, but usage of auto-redact or auto-block should really only be used if there is significant risk exposure through your harness or otherwise they are a constant source of bugs.
Recent studies by e.g. Google show a much broader range of adoption.
Also, from the trajectory of the project it doesn't look like they are significantly deoendent on outside code contributions.
You should be setting up a CLA process then ASAP if you are looking to take outside contributions.
That's up to GCP then. I'm sure OP would be willing to license pgrust to them for an appropriate price.
It looks to me like a failure of the American media landscape, that such constant revisal trends are not called out in the reporting of new numbers. That just makes them a mouthpiece of the government.
I know this may be a very poor proxy for reality, but from all the US true crime Netflix/Youtube videos I've seen over the years, it seems like there is a boatload of emergency call takers out there that are very bad at picking up the details from the caller.
I'm wondering if this is a thing similar to self driving cars, where people drastically overestimate the current baseline that humans are providing, and where automated systems are mainly not implemented due to liability reasons.
From what I've seen most auto-approvers in coding harnesses either auto-approve or auto-reject, with no middle ground of escalating the decision to the user, and breaking down the pros and cons for the decision.
Locking that down to nothing is trivial for any harness: just don't expose those to the LLM.
The tricky part is allowing access to those.
And so does academia. It's just that instead of AI and robotics, PhD students are thrown onto problems that are in large parts slightly tweaked reconfigurations of similar experiments.
Especially in chemistry, biochemistry, material sciences there is a large space of discoveries that are barely "novel" in an intellectually stimulating way, but still highly valuable that can be explored orders of magnitudes faster than is currently the case.
Once you are able to not just go on-premise, but on-device, there a boatload of other advantages than efficiency (network independence, privacy, etc.) that make it attractive.
Tiny language models are already plenty "good enough" for a lot of on-device use-cases. E.g. there are a lot of things, that can be reformulated as ranking problems, where you can apply pairwise ranking, which even small language models are good at.
EDIT: This was obviously meant to be facetious with the intent to highlight the negative effects on the general population that datacenter buildout has in the US.
According to [0], they intend to spend $18m on community projects total, out of their total $240m FY26-27 budget[1]. $330k of that go to the WikiPortraits project[2].
Almost half of their expenses still go into "keeping Wikipedia up" (+ it's companion projects). I would hardly call that deceptive.
[0]: https://meta.wikimedia.org/wiki/Grants:Regions#Wikimedia_Com...
[1]: https://meta.wikimedia.org/wiki/Wikimedia_Foundation_Annual_...
[2]: https://meta.wikimedia.org/wiki/Grants:Programs/Wikimedia_Co...
Yes, and we as "harness providers" don't want to have to hard-code integrations for each of them, and would rather provide an out-of-the box baseline in the form on an MCP server that our customers can either extend or replace with MCP servers by third-party providers. That helps us focus on the core of our offering, while also giving our customers good flexibility, which so far has been working out great.
That's exactly what standards/protocols like MCP are for. You likely nowadays also wouldn't suggest that someone builds bespoke logging/tracing integrations for each provider on the market while the good generic solution in building OTEL logging/tracing exists.
On a purely functional level, yes. However for interactive use cases, the Completions API, as provided by OpenAI or Azure, if paired with reasoning effort of any kind, provides an awful user experience, as you will have a perceived delay of 10+ seconds until the first tokens stream in.
If using other providers that are exposing their thinking traces, this is less of an issue, as they've just extended the Comletions API format to have delta events with reasoning_content.