storus

8 ms·

121.
by storus 5mo ago
Paradoxically, nicotine has some medical use in e.g. displacing viral debris and autoantibodies from nAChR (nicotinic acetylcholine receptors) due to having highest affinity to these receptors, which seems to help with (long) Covid; "s
122.
by storus 5mo ago
Can it instruct DeepSeek during an LLM call to start removing old tool calls from the context instead of waiting for the LLM call to finish if the context size approaches DeepSeek's dumb zone? Claude Code can't do that, /comp
123.
by storus 5mo ago
Not trying to be paranoid, but losing recorded history raw as it was originally reported could lead to quick AI-assisted rewrites in the archives of news outlets to fit whatever narrative of the "jour" is in fashion/that powe
124.
by storus 5mo ago
Did you think about Max-Q cards? 300W and they aren't that noisy either, 14% lower perf than non-Max-Q card.
125.
by storus 5mo ago
This is really not so clear cut as "fair use" might cover 99% of all data scrapping; you are not reproducing the originals just use them to estimate probabilistic distribution of tokens in pre-training. You are never going to get
126.
by storus 5mo ago
Agents Agents Agents!
127.
by storus 5mo ago
Winning the AI race means obsoleting their current profitable business. They might win but at what cost?
128.
by storus 5mo ago
It's 3.7-max; max was never open-weighted before. I don't see any smaller models in that tweet.
129.
by storus 5mo ago
This article had absolutely no proof of its statements and all it did was use past allegories. We are definitely missing something in our understanding of the world, like before we knew about electromagnetism or radioactivity. There are som
130.
by storus 5mo ago
How would native wolves prevent bears from coming down to human settlements?
131.
by storus 5mo ago
Soil has degraded mineral-wise so much that you won't survive even off grid and develop mineral deficiencies quickly.
132.
by storus 5mo ago
Both 10Gb/s and 8GB RAM limit come from iPhone 16 Pro chip limitations used in Neo. Next year's should have 12GB of RAM.
133.
by storus 5mo ago
And in other news some European Google Mail accounts got a notification that due to their usage patterns, they are migrating their accounts to Texas (for people who might have traveled through Texas at some point in the past).
134.
by storus 5mo ago
Depending on a task, there are already models matching Opus 4.5. Just not in everything. But you can always swap a local model for a particular task.
135.
by storus 5mo ago
It seems like M3U 512GB RAM was a unicorn we won't ever see again :( Many skipped buying it with the hopes of a 768GB-1TB M5U but it looks increasingly unlikely.
136.
by storus 5mo ago
I used to work as a VP and a part of my responsibilities was to chop up tasks to self-contained work units that can be easily assigned to random devs. This was both morally problematic for humans (i.e. EVPs forced treating human = CPU) and
137.
by storus 5mo ago
I see one profitable enterprise for AI that involves spying on everyone, managing their lives (or otherwise) tightly, automating foreign conquests and needing to make only the top decisions while delegating everything else, like a king. I c
138.
by storus 5mo ago
It's a complete takeover of technically incompetent management that feels like it can finally execute their ideas to the fullest instead of relying on those pesky swengs with their obstructions, complaints and problems. We'll soon
139.
by storus 6mo ago
That's literally how some Meta AI jobs looked a few years back - set up a few parameters, push a button, wait until training and evals are finished; repeat if. needed. $500k+/year.
140.
by storus 6mo ago
Why don't you use Opus to draft a legal letter they need to answer at the latest on some firmly set date and send it to their legal department?
141.
by storus 6mo ago
DGX Spark has actually the same compute as 5070Ti but its slower RAM and TDP brings it down to the 5070 territory.
142.
by storus 6mo ago
ARM doesn't have any initialization/boot standard enforced on phone manufacturers so everyone is doing their own and never upstreams anything. It's a general ARM problem. PC with some lucky coincidence avoided that fate and w
143.
by storus 6mo ago
DGX Spark is at the compute level of 5070. Its main issue is low memory bandwidth, i.e. it has quite fast token prefill but awful token generation. Strix Halo is just slow on every metric and used to be a cheap way to get 128GB unified RAM
144.
by storus 6mo ago
Managers now try to "extract value" quickly, leaving ruins behind them and not caring about the future as the immediate payouts allow them to stick to the "F*k you, I got mine!" paradigm.
145.
by storus 6mo ago
Schwarz seems to be obsessed with how Amazon (book seller) created AWS and they are trying to do the same... with 5 people. Also Aleph Alpha + Cohere is a Lidl work as the current CEO of the former led Lidl digital division.
146.
by storus 6mo ago
Probably usable for chat sessions, unusably slow for agentic coding.
147.
by storus 6mo ago
Oh well, I should have bought 2x 512GB RAM MacStudios, not just one :(
148.
by storus 6mo ago
4-bit quantization is not applied to all layers, some are kept 8/16-bit.
149.
by storus 6mo ago
If this runs at Opus 4.5 level for agentic coding then I don't really need any cloud models anymore.
150.
by storus 6mo ago
Step changes are often opportunities to introduce new unwanted features by default; see how countries switching to Euro experienced significant price increases on day 1. Policy makers often optimize to introduce new things the old market (i

More ›