in-silico

7 ms·

91.
by in-silico 9mo ago
I'm working on a continuous chain-of-thought reasoning architecture for generative AI models. It is similar to Meta's COCONUT. However, instead of the training forward and backwards passes of the reasoning tokens being done serial
92.
by in-silico 9mo ago
> They're going to diverge if the underlying manner in which data gets memorized and encoded, such as with RNNs, like RWKV. In the original paper ( https://arxiv.org/abs/2405.07987 ) the authors also compared the
93.
by in-silico 10mo ago
Humans are a lot dumber than you think too. They are self-replicating machines, not logical deduction engines. It says so right in the name. Expecting logic or reason or decision making is a gross mis-application.
94.
by in-silico 10mo ago
Here's a recent paper showing that models trained to generate videos develop strong geometric representations and understanding: https://arxiv.org/abs/2512.19949
95.
by in-silico 10mo ago
I want to point out this part: > A question that was interesting, but didn’t lead to a larger conclusion, was asking what actually happens when you ask a tool like ChatGPT a question. 45% think it looks up an exact answer in a database,
96.
Americans Have Mixed Views of AI – and an Appetite for Regulation(searchlightinstitute.org)
8 points by in-silico 10mo ago|2 comments
97.
by in-silico 10mo ago
> nobody has tried to generalize it for example by combining the recurrence concept with next token prediction Here you go: https://arxiv.org/abs/2502.05171
98.
by in-silico 10mo ago
> Current AI systems can give best guess statistical answer from dataset the've been fed. What about reinforcement learning? RL models don't train on an existing dataset, they try their own solutions and learn from feedback. RL
99.
The Jagged AI Frontier Is a Data Frontier(huggingface.co)
3 points by in-silico 10mo ago|0 comments
100.
by in-silico 10mo ago
It's interesting that the model generalizes to unseen participants. I was under the impression that everyone's brain patterns were different enough that the model would need to be retrained for new users. Though, I suppose if the
101.
by in-silico 11mo ago
> When that is gone and it starts doing web searches -- or it has any mechanisms that mimic actual research when it does not know something ChatGPT and Gemini (and maybe others) can already perform and cite web searches, and it vastly im
102.
by in-silico 11mo ago
The next step (and I am not claiming it's the right one) is probably "Generative UI" where the model creates website-like interfaces on the fly. Google seems to be making good progress [1] and it seems like only a matter of t
103.
by in-silico 11mo ago
Google is already doing this with Gemini: https://research.google/blog/generative-ui-a-rich-custom-vis... I don't know if/when it will actually be in consumers hands, but the tech is there.
104.
Mira Murati's Thinking Machines seeks $50B valuation in funding talks(reuters.com)
2 points by in-silico 11mo ago|0 comments
105.
by in-silico 11mo ago
It is naive to think that their end-goal is to make models that play computer games. The reason they build models like this is because playing 3D computer games is the closest digital proxy for operating a robot or similar in the real world
106.
by in-silico 11mo ago
Genie 3 (at a prototype level) achieves the goal she describes: a controllable world model with consistency and realistic physics. Its sibling Veo 3 even demonstrates some [spatial problem-solving ability]( https://video-zero-shot
107.
by in-silico 1y ago
The ideas are very interesting, but the paper is lacking in details (especially given that they defer several times to an appendix which is not present in their pdf). I don't think this could be reproduced very closely given what they
108.
by in-silico 1y ago
They basically invented LLMs as we know them (autoregressive transformers trained on web data) with the GPT 1/2/3 series of papers. They also pioneered reasoning models, which are probably the biggest breakthrough since GPT-3 on t
109.
by in-silico 1y ago
> How do we explain humans who invent new concepts Simple: they are hallucinations that turn out to be correct or useful. Ask ChatGPT to create a million new concepts that weren't in its training data and some of them are bound to b
110.
by in-silico 1y ago
Until they've undergone preference tuning and RL post-training (which is expected to start using more training compute than next-token-prediction pre-training).
111.
by in-silico 1y ago
Considering that OpenAI's model got a higher score than any of the world's best collegiate programming teams, I'd guess that a mechanical turk would not do better (even if you gave them quite a bit of time).
112.
by in-silico 1y ago
Cars going faster than humans or horses isn't very interesting these days, but it was 100+ years ago when cars were first coming on the scene. We are at that point now with AI, so a more fitting headline analogy would be "In a wor
113.
by in-silico 1y ago
The point is that up until now, humans were the best at these competitions, just like horses were the best at racing up until cars came around. The other commenter is pointing out how ridiculous it would be for someone to downplay the perfo