phyalow

7 ms·

61.
by phyalow 2y ago
In my empirical experience, its more to do with stashing the proceeds of state capture by politically connected individuals in China.
62.
by phyalow 2y ago
Investors, they announced two days ago they were conducting a $6.5B funding round. This is well timed window dressing. https://www.bloomberg.com/news/articles/2024-09-11/openai-fu...
63.
by phyalow 2y ago
https://www.lib.cam.ac.uk/using-library/your-library-members... (scroll to the bottom, any one can access).
64.
by phyalow 2y ago
RE: Your Trinity contention, I couldnt see any reference in the link to it all? Many colleges at Cambridge are directly named after Christian references. Indeed Trinity College and Trinity Hall are both ancient institutions (Newton, Hawking
65.
by phyalow 2y ago
Gaussian Process Regression (a form of Bayesian Optimisation to try and get to the right "answer"/parameter space sooner) - explained in some context here... https://brendanhasz.github.io/2019/03/28&
66.
by phyalow 2y ago
Agree. (D)NNs have a powerful but somewhat loose inductive bias. They're great at capturing surface-level complexity but often miss the deeper compositional structure. This looseness, in my opinion, stems from a combination of factors:
67.
by phyalow 2y ago
It's been a popular topic for the last couple of weeks on /r/claudeAI/. I work via the API (console.anthropic.com/workbench) which is still showing a static model number and has less safety filtering applied to it t
68.
by phyalow 2y ago
Genuine question, whats the difference between your startup and just calling the below code with a different model on a cloud machine, other than some ML/Dev OP's engineer not knowing what they are doing...? model = get_model(
69.
by phyalow 2y ago
Thats because GPT products are in a different (much worse) universe to Anthropics Sonnet/Opus, they are truly phenomenal. Give Anthropic a shot (its even better via the API console.anthropic.com/workbench). OpenAI is yesterdays ne
70.
by phyalow 2y ago
I feel like its just getting at the main ideas of https://en.wikipedia.org/wiki/Statistical_mechanics (although I am technically lay in this field).
71.
by phyalow 2y ago
Thats true, for OpenAI's/FAIR's LLM's sure... Claude Opus or Sonnet 3.5? Not in my experience...
72.
by phyalow 2y ago
Empirically while you wait for yet another print on arxiv, I use the Sonnet 3.5 API every day to reason problems, iterate ideas and write highly optimised python and things like CUDA kernals. There is some degree of branching, guiding and c
73.
by phyalow 2y ago
Absolutely.
74.
by phyalow 2y ago
https://www.youtube.com/watch?v=A4TXAaqmj0E This video of Kim is a classic (I think it came out origianlly c. 2009). The last shot of his table with a stack of Gold Bars in the middle is very revealing (or maybe moreso is h
75.
by phyalow 2y ago
I disagree, he has engaged in a deliberate media/public personality building campaign over the last 12 years to make the public sympathetic to him and his plight. >>I do not think he's a criminal "He was arrested in 199
76.
by phyalow 2y ago
Correct me if im wrong, this is just a question originating from my own curiosity, but could you not apply the same or similar techniques to classical NN architectures? I dont see exactly how the differentiability of the architecture matter
77.
by phyalow 2y ago
Experimentation for Engineers: From A/B Testing to Bayesian Optimization. by David Sweet This book is really great, and I highly recommend it, it goes broader than A/B, but covers everything quite well from a first principles pers
78.
by phyalow 2y ago
I think in more esoteric languages it wont work as well. Python, C++ it is excellent, suprisingly its Rust is also pretty damn good. (I am not a paid shiller, just in awe of what Sonnet 3.5 + Opus can do)
79.
by phyalow 2y ago
> Figuring out how to compact the total measure of fitness into a single float as the simulation progresses is non trivial. I've found that dynamically adapting the fitness function and simulation parameters over time is essential.
80.
by phyalow 2y ago
Reservoir computers; https://en.m.wikipedia.org/wiki/Reservoir_computing
81.
by phyalow 2y ago
Probably something like this: https://en.wikipedia.org/wiki/Creatures_(1996_video_game) Many others mentioned here: https://en.wikipedia.org/wiki/Life_simulation_game
82.
by phyalow 2y ago
The inevitable march of entropy.
83.
by phyalow 2y ago
Have you used Claude Opus or Sonnet 3.5 by anychance?
84.
by phyalow 2y ago
its called running and editing notebooks with vscode...
85.
by phyalow 2y ago
As a complete rebuttal, https://caityjohnstone.medium.com/assange-smear-2-hes-a-rapi... Or more broadly any in this great series of posts by Caitlan Johnstone: https://medium.com/search?q=Assange+Smear+caitl
86.
by phyalow 2y ago
I have this one (basic kit): https://www.etsy.com/listing/1677292928/comfort-system-for-t... its really great, its like wearing thick rimmed glasses instead of a scuba mask. You take the light seal off, and the vi
87.
by phyalow 2y ago
I hadnt been for a month or so (picked it up at launch), but the new beta release of Vision OS 2 looks feels like a massive quality of life improvement. Foveation and implied resolution seem to be massively improved, framerates are much hig
88.
by phyalow 2y ago
Dont knock it if works. Simple units still have nice properties and dont require as much compute to solve for.
89.
by phyalow 2y ago
Im willing to trade a barely used AVP for ££
90.
by phyalow 2y ago
New Zealand too. I totally agree. There is something utterly rotten with the seniorage abilities of commercial banks and the multigenerational distortions it has created by virtue of leveraged (mortgage) lending.

More ›