storus

15 ms·

391.
by storus 10mo ago
Disabling iCloud seems like a gift. I wish I could just get rid of it all without any subsequent nagging every time I update/upgrade macOS.
392.
by storus 10mo ago
Something similar happened to me, but with Lufthansa. Canceled my flight 1 day in advance and told me to take a hike, didn't even bother to find/recommend another flight. Germany has really deteriorated, it's no longer matchi
393.
by storus 10mo ago
It was about this use case: https://blog.exolabs.net/nvidia-dgx-spark/
394.
by storus 10mo ago
Is there any way to connect DGX Sparks to this via USB4? Right now only 10GbE can be used despite both Spark and MacStudio having vastly faster options.
395.
by storus 10mo ago
They probably don't know why it was de-indexed either, likely a bunch of unexplainable ML models flagged it.
396.
by storus 11mo ago
Apple has missed the train, from high-end computing they ended up as a low-end ML company (the most expensive MacStudio is way cheaper than a proper Nvidia card), and now some analyst is seeing virtue in that? They completely missed out on
397.
by storus 11mo ago
Another article that wants to impose something on a tech we don't really understand and that works the way it works by some happy accident. Instead of pushing the tech as far as we can, learning how to utilize it and what its limitatio
398.
by storus 11mo ago
GT used to have a collaboration with Meta's FAIR and one could do final projects in the DL and NLP classes on recent research topics from Meta AI; some people ended up at top conferences with their work. Not sure how it's going to
399.
by storus 11mo ago
UTexas seems to be crushing it in the ML/AI space as they offer far more recent courses with deeper topics; for everything else OMSCS is probably a better choice even though it has a relentless pace of busywork making even easy classes
400.
by storus 11mo ago
Udacity recently released their own MS AI though with shaky accreditation via Malta's MFHEA which was recently rejected by EQAR, making it unaccredited abroad.
401.
by storus 11mo ago
Linear attention is really bad, it's only good for benchmaxing but it leads to a loss of valuable granularity, which can be felt in the latest DeepSeek randomly forgetting/ignoring/correcting explicitly stated facts in the pr
402.
by storus 11mo ago
HMD under NOKIA brand went almost out of business due to adding notches to their phones, now Jolla is doing the same mistake. Only Apple could get away with it. At least they aren't shipping a 720p display anymore. Why didn't they
403.
by storus 11mo ago
Dunno, DeepSeek on MacStudio doesn't feel much slower than when using it directly on deepseek.com; 6t/s is still around 24 characters per second which is faster than many people could read. I also have 6000 Pro but you won't
404.
by storus 11mo ago
I addressed both points - I mentioned you can offload token prefill (the slow part, 9t/s) to DGX Spark. Token generation is at 6t/s which is acceptable.
405.
by storus 11mo ago
Didn't that "good enough" strategy fail with Global Foundries? They "good enoughed" themselves to irrelevance.
406.
by storus 11mo ago
I run it all the time, token generation is pretty good. Just large contexts are slow but you can hook a DGX Spark via Exo Labs stack and outsource token prefill to it. Upcoming M5 Ultra should be faster than Spark in token prefill as well.
407.
by storus 11mo ago
I suspect they killed it a bit earlier but in different regions and only now it arrived to the US; I went to Croatia in September and couldn't get Netflix to stream from any of my devices onto the hotel room's TV and had to instea
408.
by storus 11mo ago
They can easily LoRA-finetune each model based on user preferences expressed in the past conversations. That would improve accuracy compared to Google's ad targeting by orders of magnitude.
409.
by storus 11mo ago
You can find devices on ebay that should detect cameras by using a polarized glass and LED lights as cameras tend to reflect back well and you'd see a shining spot where sensor is.
410.
by storus 11mo ago
Melanin is used as a solar panel, capturing gamma rays and then passing the resulting consistent flow of excited electrons over to the Krebs cycle with the help of CO2; like in photosynthesis. Humans run on electrons resulting from Krebs cy
411.
by storus 11mo ago
More like 900% right now.
412.
by storus 11mo ago
If Google won, it would cannibalize its current ad-driven business and replace it with something that is extremely expensive to run and difficult to make profit from. A Pyrrhic win essentially.
413.
by storus 11mo ago
It's also covered by CS329H.
414.
by storus 11mo ago
Those don't have DPO/GRPO which arguably made some parts of RL obsolete.
415.
by storus 11mo ago
Thanks, that looks very promising!
416.
by storus 11mo ago
RL is extremely brittle, it's often difficult to make it converge. Even Stanford folks admit that. Are there any solutions for this?
417.
by storus 11mo ago
I thought that was an ancient issue with Samsung 740? I had that one and it was slowly losing speed when unpowered due to an accumulation of errors and rewriting the individual sectors once for the whole drive made it work fine for a year.
418.
by storus 11mo ago
What are the current ways to minimize context usage when streaming with multiple tool calls? I can offload some stuff to tools themselves, i.e. they wrap some LLM doing heavy lifting like going through a 200k-token-long markdown and return
419.
by storus 11mo ago
If you consider surveillance capitalism and dark pattern nudges a good thing, then sure. Gemini has the potential to obliterate their current business model completely so I wouldn't consider that "waking up".
420.
by storus 11mo ago
It was somewhere on HN in the past few weeks, a few folks mentioned they can't boot some systems anymore.

More ›