Philpax

31 ms·

781.
by Philpax 2y ago
I don't know if that's the correct interpretation? I would say they degraded the UX because they aren't hoovering up enough data to make their solution work properly, unlike Google and Amazon. They've done a better job
782.
by Philpax 2y ago
Yes, the spatial videos that iPhones can record are just stereoscopic video. This hasn't been cloned by other manufacturers because there have been very few viewing devices that they own, but that might change with Android XR.
783.
by Philpax 2y ago
To be fair, this is graphical, but it's not real-time, not performant, and while it technically takes input, it'll be a while before you see it reflected in the output.
784.
by Philpax 2y ago
All good! It is what it is :)
785.
by Philpax 2y ago
Fixed :>
786.
MichiganTypeScript: A WebAssembly runtime implemented in TypeScript types(github.com)
88 points by Philpax 2y ago|8 comments
787.
Anthropic: Forecasting rare language model behaviors(anthropic.com)
2 points by Philpax 2y ago|0 comments
788.
Recursive Improvement: AI Singularity or Just Benchmark Saturation?(timkellogg.me)
3 points by Philpax 2y ago|0 comments
789.
AI progress is about to speed up(epochai.substack.com)
2 points by Philpax 2y ago|0 comments
790.
by Philpax 2y ago
It's run by Anthropic! https://x.com/AnthropicAI/status/1894419011569344978
791.
by Philpax 2y ago
This is truly tremendous to watch. Eleven years from TPP, and we're watching the current best-in-class AI try its best at the same. Who'll get there first, the historical gestalt of Twitch users or the just-shy-of-10^26 FLOPS [0]
792.
<think> </think> QwQ-Max-Preview(qwenlm.github.io)
6 points by Philpax 2y ago|0 comments
793.
by Philpax 2y ago
Aha, there it is - terrific, thank you :> Yes, I'm quite the eclectic kind!
794.
by Philpax 2y ago
New benchmark for good posting, I'll take it!
795.
by Philpax 2y ago
Eh, to some extent - there's still a pretty significant cost to actually running inference for those models. For example, no consumer can run DeepSeek v3/r1 - that requires tens, possibly hundreds, of thousands of dollars of hardw
796.
by Philpax 2y ago
The fundamental innovation is training the model to reason through reinforcement learning; you can train existing models with traces from these reasoning models to get you within the same ballpark, but taking it further requires you to do R
797.
by Philpax 2y ago
Seems broken? Getting > An error occurred in the Server Components render. The specific message is omitted in production builds to avoid leaking sensitive details. A digest property is included on this error instance which may provide ad
798.
Claude 3.7 Sonnet System Card [pdf](assets.anthropic.com)
3 points by Philpax 2y ago|0 comments
799.
anthropic.claude-3-7-sonnet-20250219-v1:0(twitter.com)
8 points by Philpax 2y ago|1 comments
800.
by Philpax 2y ago
Looks like the new Claude should be on its way shortly (data-mined from website/app).
801.
by Philpax 2y ago
Fine for you, perhaps, but not fine for everybody.
802.
by Philpax 2y ago
"Trivially parseable by machines" is not mutually exclusive with "Trivially parseable by humans". JSON is both.
803.
by Philpax 2y ago
That's the result of the competition, but what I want is a demonstration of the actual project: in an uncongested area, take a video of someone walking behind a wall and demonstrate that you can still locate them.
804.
by Philpax 2y ago
I understand that this was meant to be a retelling of how the build process went and what happened afterwards, but it seems quite light on the actual results? I would have liked to see a video demonstrating how well it lets the user see thr
805.
by Philpax 2y ago
Watching the speedrunner improve, watching them discover new techniques, the discussion they have with their audience, etc. Speedrunning, ironically, is not just about the destination: it's about the (often public!) journey the speedru
806.
by Philpax 2y ago
> I couldn't care less what the authors personal beliefs and political leanings are, or who they offended on IRC or social media. I just don't really want to use or support software by people who, at best, think it's appro
807.
by Philpax 2y ago
This doesn't seem to really address synthetic data, let alone RL-based reasoning.
808.
by Philpax 2y ago
Ramanujan is another great example of divine inspiration.
809.
by Philpax 2y ago
Tom Scott video about this: https://www.youtube.com/watch?v=1bO8zEaSuWg
810.
by Philpax 2y ago
Sounds like what's described in Manna: https://marshallbrain.com/manna1

More ›