storus

13 ms·

301.
by storus 8mo ago
Didn't the same people who wrote Delphi write also C# and .NET? When I first saw .NET it felt very much like VCL/CLX. And then came TypeScript from the same guy.
302.
by storus 8mo ago
The point is that LLMs can only replicate what existed somewhere, they aren't able to invent new things. Once humans lose their edge, there won't be any AI-driven progress, just a remix of existing stuff. That was the hollowing ou
303.
by storus 8mo ago
It will lead to hollowing out of the substance everywhere. The constant march to more abstraction and simplicity will inevitably end up with AI doing all the work and nobody understanding what is going on underneath, turning technology into
304.
by storus 8mo ago
Can somebody tell me what's behind Thiel's obsession with Antichrist?
305.
by storus 8mo ago
Just add a DGX Spark for token prefill and stream it to M3 using Exo. M5 Ultra should have about the same compute as DGX Spark for FP4 and you don't have to wait until Apple releases it. Also, a 128GB "appliance" like that is
306.
by storus 8mo ago
Current LLMs use special tokens for tool calls and are thoroughly trained for that, nearing almost 100% correctness these days, allowing multiple tool calls per single LLM response. That's hard to beat with custom tool calls. Even olde
307.
by storus 8mo ago
Anyone using OpenClaw to manage a bunch of coding agents so that you only set the high-level vision and leave all the prompting, testing, debugging, forking to agents? If yes, how did you glue it all together? Are you using local models? Wh
308.
by storus 8mo ago
This. It's awful to wait 15 minutes for M3 Ultra to start generating tokens when your coding agent has 100k+ tokens in its context. This can be partially offset by adding DGX Spark to accelerate this phase. M5 Ultra should be like DGX
309.
by storus 8mo ago
Even grad level?
310.
by storus 8mo ago
SEPA gets blocked immediately when you try to buy something expensive, like a top-end graphics card (8k+).
311.
by storus 8mo ago
Touchscreen insanity and even rounder corners than Tahoe. I guess I am not their target customer.
312.
by storus 8mo ago
There was this sentence in the article: "...he realized that nonuniformly elliptic PDEs that seem well behaved can have irregular solutions even when they satisfy the condition Schauder had identified"
313.
by storus 8mo ago
I use Discord to talk to university students (top 10 in CS) and it only works with university email. I am wondering if I am going to be treated as <13 from now on as well or if they waive it in our case.
314.
by storus 8mo ago
How can they be sure they got it right when the previous canonical knowledge (Schauder) turned out to be false?
315.
by storus 8mo ago
The "marathon of sprints" paradigm is now everywhere and AI is turning it to 120%. I am not sure how many devs can keep sprinting all the time without any rest. AI maybe can help but it tends to go off-rails quickly when not super
316.
by storus 8mo ago
Wouldn't Omega-3 and vitamin B2 together be a great prophylaxis for most neuro-degenerative conditions due to repolarizing microglia?
317.
by storus 8mo ago
It can't be uninstalled on Sonoma I am using.
318.
by storus 9mo ago
I never use Apple News but they often pop up among the apps that are using significant energy. I am wondering what does it really do on the background.
319.
by storus 9mo ago
With BM25 which has a far worse/non-generalizable performance than sparse embeddings Pinecone supports. Moreover you get a latency hit from RRF that makes it challenging to use for e.g. real-time multimodal chat agents.
320.
by storus 9mo ago
Now this is going to be interesting to watch to see if the finance bros financing this AI wave to get rid of SW engineers will keep financing getting rid of their own.
321.
by storus 9mo ago
BM25 is quite bad and needs to be retrained for each corpus anew. SPLADEv2 is much better and there are even better sparse embeddings these days.
322.
by storus 9mo ago
Pinecone allows hybrid search, merging dense and sparse vector embeddings that Postgres can't do AFAIK. That results in ~10% worse retrieval scores which might be the difference between making it in the business or not.
323.
by storus 9mo ago
Now this is fairly "easy" as there are multitude of implementations/specs all over the Internet. How about trying to design a new language that is unquestionably better/safer/faster for low-level system programming
324.
by storus 9mo ago
Wasn't the ivermectin hypothesis based on it blocking NS3 helicase preventing double-stranded viral RNA from being unwound after replication? Paxlovid targets a few steps earlier by blocking 3CL protease that chops viral RNA into funct
325.
by storus 9mo ago
I had two M3 Max 128GB RAM/8TB SSD that had GPU issues (the second one was likely a refurb so probably returned by somebody else) and am now waiting for the third one. M3 Ultra 512GB RAM was rock solid so far though.
326.
by storus 9mo ago
OK, so you are essentially using sub-agents as summarizing tools of the main agent, something you could implement by specialized tools that wrap independent LLM calls with the prompts of your sub-agents.
327.
by storus 9mo ago
I still have 2x NVLinked A6000 and they aren't that bad compared to a single RTX 6000 Pro.
328.
by storus 9mo ago
Aren't layers basically doing n^k attention? The attention block is n^2 because it allows 1 number per input/output pair. But nothing prevents you from stacking these on top of each other and get k-th order of "attentioness&q
329.
by storus 9mo ago
That's why you can use latest open coding models locally that reportedly reached the performance of Sonet-4.5 so almost SOTA. And then you can think of tricks like I mentioned above to directly manipulate GPU RAM for context cleanup wh
330.
by storus 9mo ago
Yeah but this seems like a bolt-on and not something they train their model to understand at the token level like how they do tool calls. Maybe Anthropic has a token-level skills support (e.g. <SKILL_START>skill prompt<SKILL_END&gt

More ›