dfsegoat
Joined 11 years ago
Karma
2,416
Submissions
637
Member since
Sep 11
Recent submissions
Comment · 9 months ago
Comment · 10 months ago
Highly recommend using agent based hooks for things like `[review & test]`.
At a basic level, they work akin to git-hooks, but they fire up a whole new context whenever certain events trigger (E.g. another agent finishes implementing changes) - and that hook instance is independent of the implementation context (which is great, as for the review case it is a semi-independent reviewer).
Comment · 10 months ago
It seems that perhaps neither are inherently 'good models'? What would an ideal alternative look like?
Comment · 10 months ago
> it’s more performant than Claude Opus 4.5 or GPT 5.2 extra high
...and all of that done without any GPUs as far as i know! [1]
[1] - https://www.uncoveralpha.com/p/the-chip-made-for-the-ai-infe...
(tldr: afaik Google trained Gemini 3 entirely on tensor processing units - TPUs)
Comment · 10 months ago
they 100% are unless you provide a RUBRIC / basically make it ordinal.
"Return a score of 0.0 if ...., Return a score of 0.5 if .... , Return a score of 1.0 if ..."
Comment · 1 year ago
Very similar usage duration / window for me (US Pacific) - no availability issues.
Comment · 1 year ago
To put some numbers to trying to develop a single therapy (where candidates etc. will fail as you try them)
- Plan to sink $180-500M+ just in R&D
- Factor in failures, regulatory, clinical, recruitment, phase 1/2 trials and you arrive very quickly around $1.3-2.1 BILLION USD per therapy approved.
...there is a 90% chance that you will spend that $1B+ - and it will fail completely.
https://www.nature.com/articles/d41573-020-00043-x
https://greenfieldchemical.com/2023/08/10/the-staggering-cos...
Comment · 1 year ago
fwiw towards your theory, I believe that the US Govt actually considers cloud providers - by way of specific services offered "dual use" systems for mil or civil use.
E.g. you will find references in AWS docs to Bureau of Industry/Security rulings.
Comment · 1 year ago
OTP protected flow verification
Comment · 1 year ago
Really good call out re: email and other 'side-flows' - hopefully there is integration with something like Mailosaur.
Comment · 1 year ago
> nor are they...able to mutate themselves to gain new information on the fly
See "Self-Adapting Language Models" from a group out of MIT recently which really gets at exactly that.
Comment · 1 year ago
...not monetized yet: Can't find the post, but a prev. HN post had a link to an article showing that OpenAI had hired someone from Meta's ad service leadership - so I took that to mean it's a matter of time.
edit: believe it was Fidji Simo et al.
https://www.pymnts.com/artificial-intelligence-2/2025/openai...
Comment · 1 year ago
Ahh thank you for catching - apologies I didn't see this in time to add a correction.
Comment · 1 year ago
Related:
KingJamesProgramming [1] - a mashup of Knuth books [2] and the King James Bible with a Markov chain, is still one of my favorite reads for a laugh.
It was also the first place I was really exposed to probabilistic-based methods for generation of novel content.
You get gems like:
"they smote the city with the edge of the sword. 22:20 And one of his main motivations was the high cost of proprietary software houses."
and
"37:29 The righteous shall inherit the land, and leave it for an inheritance unto the children of Gad according to the number of steps that is linear in b."
Comment · 1 year ago
I'm typically pretty critical of PM oriented pieces, but I found this to be a decent overview of how to reason about building these systems from first principles + some of the non-tech pain points + how to address them.
Can you clarify what you mean? I am not sure I follow.