riku_iki

9 ms·

151.
by riku_iki 11mo ago
It's all small products which didn't receive traction.
152.
by riku_iki 11mo ago
> the real-world stupidity of these models such as his example of a coding agent toggling between generating bug #1 vs bug #2, which in fact largely explains why the current economic and visible impact is much less than if the "AI i
153.
by riku_iki 11mo ago
it could be results of corps focusing resources on IMO in PR wars, and results is not as generalizable outside this niche.
154.
by riku_iki 11mo ago
> "ordinary people got a bit of gold, selfish vulture capitalists butchered the goose some ordinary people who worked hard and made it happened: founders and early workers, usually rewarded very well
155.
by riku_iki 11mo ago
He personally unlikely will make buck off this. Going public usually means unlocking funding, so they could grow business, build more services, compete with say OVH or if some hyperscaler will decide to step into this niche.
156.
by riku_iki 11mo ago
with fineprint somewhere on page #67, that there are exceptions.
157.
by riku_iki 11mo ago
they have two sets: - semi-private, which they use to test proprietary models and which could be leaked -private: used to test downloadable open source models. ARG-AGI prize itself is for open source models.
158.
by riku_iki 11mo ago
> internal team to create an ARC replica, covering very similar puzzles they can target benchmark directly, not just replica. If google or OAI are bad actors, they already have benchmark data from previous runs.
159.
by riku_iki 11mo ago
> Why in the world do people choose Mongo over Postgres? Postgrtes distributed story is more complicated.
160.
by riku_iki 11mo ago
> This is not the point though could you expand what is the point? That authors opinion without much justification is that this is not reasoning?
161.
by riku_iki 11mo ago
Even in your example (which is obviously not correct representation of prolog), that code will work X orders magnitude faster and with 100% reliability compared to much more inferior LLM reasoning capabilities.
162.
by riku_iki 1y ago
Playing cards right means do not upset superrich. Robbing taxpayers and pension funds is Ok.
163.
by riku_iki 1y ago
CEOs who fraud superrich totally go to jail.
164.
by riku_iki 1y ago
I assume training set components have also priorities, low priority data goes to training very few times at the beginning of pretraining, while higher priority data is trained on multiple times until the end.
165.
by riku_iki 1y ago
Additionally, it would be interesting to know if there is dynamics in opposite directions, US corps (oai, xai) can now incorporate Chinese models into their core models as one/several expert towers.
166.
by riku_iki 1y ago
> That's not the same as starting with open model weights and training on those - until recently (gpt-oss) OpenAI didn't release their model weights. there was obviously llama.
167.
by riku_iki 1y ago
You either don't know which training data was used for say chatgpt oss, or training data can be included into some open dataset like pile or similar. I think this test is very unreliable, and even if someone come to such conclusion, no
168.
by riku_iki 1y ago
not sure what kind of evidence it could be..
169.
by riku_iki 1y ago
its moe, each expert tower can be branched from some smaller model.
170.
by riku_iki 1y ago
> How do the Chinese train these models if they don't have access to the GPUs to train them? they may be taking some western models: llama, chatgpt-oss, gemma, mistral, etc, and do postraining, which required way less resources.
171.
by riku_iki 1y ago
> If there are any such loan guarantees (which is very likely), they'll be for AI investments in the US in general, so, for example, Anthropic and Google will also benefit (and probably Nvidia, AMD, Intel, Oracle, Amazon, Meta, etc)
172.
by riku_iki 1y ago
It takes 5 sec for me to send CV, but here you ask for 20 mins commitment from my side.
173.
by riku_iki 1y ago
> Now compare that to a 20-minute screening interview where most of that pressure is gone. You can do it whenever you want to. how its gone? Candidate is still being judged, but now by unknown potential AI judge without understanding how
174.
by riku_iki 1y ago
But we don't know if hiring person spends quality time on looking at all these recordings and transcripts, or just do some superficial keyword matching thing or AI signal.
175.
by riku_iki 1y ago
> do full text search. And if you put all that together, you have a complex Postgres deployment. search is also just extension? So, its a strong point: you have one self-contained server with simple installation/maintenance story.
176.
by riku_iki 1y ago
AlloyDb is not opensource, so it is kinda different niche.
177.
by riku_iki 1y ago
> Is there a comprehensive leaderboard like ClickBench clickbench has 100m rows of data only, which makes it not comprehensive benchmark at all.
178.
by riku_iki 1y ago
I am not interested in this discussion anymore. Bye.
179.
by riku_iki 1y ago
> only looking at crappy old model let me repeat this: it was newly trained specialized model other rants are ignored.
180.
by riku_iki 1y ago
> You cannot use crappy 2-year-old models with chain-of-thought to make inferences about frontier reasoning models that were released less than a year ago. the idea is to train new specialized model, which could specifically demonstrate

More ›