imjonse

8 ms·

91.
by imjonse 1y ago
LLMs should not replace most specialized solutions but they still can help do a large part of the tasks those specialized solutions are used for today.
92.
by imjonse 1y ago
Sad to see AGI being implicitly equated with the most powerful weapon that will help the owners rule over resources instead of a scientific breakthrough that will help solve humanity's biggest problems.
93.
by imjonse 1y ago
Wish it was only true in tech...
94.
by imjonse 1y ago
The author does not seem to go into details, so I am curious what actually surprising conclusions can be drawn from wearing one of these devices? Croissants and muffins being unhealthy should be no surprise. I am more interested in findings
95.
by imjonse 1y ago
Hi. It is unclear from the README whether the free limits apply also when there's an API key found in the environment - not explicitly set for this tool - and there is no login requirement.
96.
by imjonse 1y ago
Tech support is clearly very important but I have a hard time believing there wasn't a great amount of lobbying involved as well.
97.
by imjonse 1y ago
"we don’t keep logs of who EVERYONE is messaging" just selected people then?
98.
by imjonse 1y ago
the main site is confusing indeed with all those leaderboards, but follow the discord and resources links for the actual learning material.
99.
by imjonse 1y ago
These should keep you busy for months: https://www.gpumode.com/ resources and discord community Book: Programming massively parallel processors nvidia cuda docs are very comprehensive too https://github.com/
100.
by imjonse 1y ago
"Across a decade working at hypergrowth tech companies like Meta and Pinterest, I constantly struggled with procrastination [...] I was not making progress on the things that mattered." Maybe unless one can really convince themsel
101.
by imjonse 1y ago
amusing, but true.
102.
by imjonse 1y ago
You can use a single app, and it is probably the best way to go for the majority of projects - definitely the case for simple ones.
103.
by imjonse 2y ago
Oh good, they finally realized GraphQL was holding them back.
104.
by imjonse 2y ago
Security/cryptographic strength are indeed relative, they depend on the 'threat model' being used.
105.
by imjonse 2y ago
According to the article this software is used for all major sporting events.
106.
by imjonse 2y ago
Good question. That was an issue with tanh as activation function, and before residual connections and normalization layers. Tanh as a normalization but with other activations and residual present apparently is ok.
107.
by imjonse 2y ago
Some of their research had already broken out into mainstream, DDIM at least was their paper and probably others too in the diffusion domain.
108.
by imjonse 2y ago
yes, biased against knowing 'the best way do write software' and applying it regardless of what the current requirements and constraints are. And arguing for their position by sending people links to Uncle Bob videos for 'enl
109.
by imjonse 2y ago
> So if someone is 60+ year old, chances are that most of his work has never been open sourced, John Ousterhout is 70 years old and one of the open source pioneers. We don't know what Uncle Bob shipped or did not ship but his friend
110.
by imjonse 2y ago
I am biased ( a former coworker was an Uncle Bob fan, and was bent on doing everything by the book, with layers of abstraction, patterns, hexagonal architecture, lots of unit tests, no cutting corners, even as we did not know what exactly w
111.
by imjonse 2y ago
There is a large variety between perfect code and code people usually complain about. Not only weak engineers complain about crappy code and stupid decisions.
112.
by imjonse 2y ago
Something like this, using Simon Willison's llm cli, will give you a better experience than such generated articles unless your goal is explicitly to read blogs. wget https://arxiv.org/pdf/2501.00663 -O - | pdftot
113.
by imjonse 2y ago
Is it established whether GRPO is essential for this to work as it does, or could other RLHF-class methods provide similar results? My initial (possibly mistaken) impression was that GRPO was one of ways of mitigating the lack of enormous h
114.
by imjonse 2y ago
Exclusive contracts with the defense industry or similar deals?
115.
by imjonse 2y ago
'In recent years, innovative AI products that didn’t build their own models were derided as low-tech “GPT wrappers.” ' The ones derided were those claiming to be 'open-source XY' while being a standard tailwind template
116.
by imjonse 2y ago
"as we get closer to achieving AGI, we believe that trending more towards individual empowerment is important; the other likely path we can see is AI being used by authoritarian governments to control their population through mass surv
117.
by imjonse 2y ago
The paper predates Qwen2 and R1, this work is probably a year old.
118.
by imjonse 2y ago
> ORMs are the devil in all languages and all implementations. Just write the damn SQL It depends on what you're writing. I've seen enough projects writing raw SQL because of aversion to ORMs being bogged down in reinventing a
119.
by imjonse 2y ago
dynamic typing incurs runtime overhead
120.
by imjonse 2y ago
if there is so much value for a small group, it is likely those are not simple inferences but of the new expensive kind with very long CoT chains and reasoning. So not cheap and it is exactly this trend towards inference time compute that m

More ›