llm_nerd

40 ms·

721.
by llm_nerd 1y ago
The Swift compiler does lifetime and ownership analysis to eliminate many/most ARC overhead, beyond when things are truly shared between threads and the like. https://g.co/gemini/share/51670084cd0f - lame, bu
722.
by llm_nerd 1y ago
This whole story sounds ridiculous. And I don't think he's clueless, but instead the guy wanted to bring attention to his bizarre "B2B + AI Community, Events, Leads", so setting up such a predictable footgun scenario see
723.
by llm_nerd 1y ago
Modern OCR is using machine learning technologies, including ViT and precisely the same models and technologies used in the linked solution. I mean, if their comparison was with OCR from 2002, sure, but they're comparing against modern
724.
by llm_nerd 1y ago
Strangely the linked marketing text repeatedly comments regarding OCR errors (I counted at least 4 separate instances), which is extremely weird because such a visual RAG suffers precisely the same problem . It is such a weird thing to rep
725.
by llm_nerd 1y ago
FWIW, there is a very high probability that Google's $20B yearly payment to Apple is going to vanish, pending a current trial. Safari is actually a pretty great browser, both technically and from a user perspective, and the complaints
726.
by llm_nerd 1y ago
It's a fairly obvious truth. Chrome has a 70% browser marketshare, with the only appreciable alternative being Safari. If Chrome (the actual Chrome and not just a skinned Safari) were allowed on Apple's platforms, overnight a crap
727.
by llm_nerd 1y ago
This is a fun read, however- "Life insurers can predict when you'll die with about 98% accuracy." This conclusion isn't supported by the linked document. The document instead is talking about expected vs actual deaths am
728.
by llm_nerd 1y ago
There are a huge number of reasons for large scale systems. Batching sizes when hitting MoE systems (which are basically all LLMs now) leading to routing variations. Consecutive submissions could be routed to entirely different hardware, so
729.
by llm_nerd 1y ago
>That's why vendors don't offer seed parameters accompanied by a promise that it will result in deterministic results - because that's a promise they cannot keep. They absolutely can keep such a promise, which anyone who h
730.
by llm_nerd 1y ago
That non-deterministic claim, along with the rather ludicrous claim that this is all just some accidental self-awareness of the model or something (rather than Elon clearly and obviously sticking his fat fingers into the machine), make the
731.
by llm_nerd 1y ago
>Start-ups avoid them like the plague because they're so damn expensive While in a way it's just a corollary on the expensive bit, the license compliance of the same becomes such a monumental hassle as well, and is just an en
732.
by llm_nerd 1y ago
Sure thing.
733.
by llm_nerd 1y ago
I was being rhetorical. The R in RAG is filtering augmentation data (the A) for things that might or might not be related to the query. Including everything is just a lazy form of RAG -- the rhetorical SELECT *. >and adding that as a sys
734.
by llm_nerd 1y ago
In this case their Retrieval stage is "SELECT *", basically, so sure I'm being loose with the terminology, but otherwise it's just a non-selective RAG. Okay ..AG. RAG is selecting pertinent information to supply to the L
735.
by llm_nerd 1y ago
What you described is RAG. Inefficient RAG, but still RAG. And it's inefficient in two ways- -you're using extra tokens for every query, which adds up. -you're making the LLM less precise by overloading it with potentially ir
736.
by llm_nerd 1y ago
I'm not advocating against using vectorization/SIMD. Of course people should eek every bit of performance out. The point is that the average developers job in achieving this is not knowing every SIMD primitive for every platform.
737.
by llm_nerd 1y ago
For sure there are obviously developers building those computation libraries like numpy, compilers, R, and so on. These people exist and are grinding out great code and abstractions for the rest of us to use, and many of them are regulars o
738.
by llm_nerd 1y ago
Neat tool. It is interesting how often SIMD stuff is discussed on here. Are people really directly dealing with SIMD calls a lot? I get the draw -- this sort of to-the-metal hyper-optimization is legitimately fun and intellectually rewardin
739.
by llm_nerd 1y ago
As a note on Microsoft's obnoxious Copilot push, I too got the "Your 365 subscription price is increasing because we're forcing AI on you". Only when I went to cancel[1], suddenly they made me aware that there was a &quo
740.
by llm_nerd 1y ago
As one aside on this, that Dr. Layne Norton video has drawn a lot of comments by people who declare that while CAC scores are rising dangerously in the LMHR Keto group --- exactly the opposite to what the LMHR and Keto communities predicted
741.
by llm_nerd 1y ago
Statin uses translate into less death, without any medical question or doubt. That particular systematic review is notorious for being utterly nonsensical. It's also notable that Maryanne Demasi is of...uh...of dubious professional cre
742.
by llm_nerd 1y ago
>This is a naive perspective. Okay. SR&ED had 22,758 applications last year. Software development only accounted for 40% of it. So 9000 applications from software dev firms, the majority being very small firms. That is a tiny, tiny m
743.
by llm_nerd 1y ago
Canada is much more socialist, in your take, so it has more programs for corporations and private enterprise? Huh? This is nonsensical. Further, it's incredibly difficult to quantify countries on this purported socialism scale. Sure, C
744.
by llm_nerd 1y ago
It's 35% of eligible spend on up to $3 million, and 15% above that (15% and 15% if the corporation is not Canadian). Further, most software development simply doesn't qualify- https://www.canada.ca/en/revenue-
745.
by llm_nerd 1y ago
Basing anything on Google's pricing is folly. Quite recently Google offered several of their preview models at a price of $0.00. Because they were the underdog. Everyone was talking about ChatGPT, or maybe Anthropic. Then Deepseek. Goo
746.
by llm_nerd 1y ago
Yet another B of L stellar job report just dropped. Once again "higher than expected" jobs. Anyone want to take bets on how the revisions will go on this?
747.
by llm_nerd 1y ago
What do you think the "reason" is? Several other comments have made self-contradictory claims. NEVER has it consistently over-estimated jobs as it has, in the history of reporting. Look, yet another stellar jobs report from the B
748.
by llm_nerd 1y ago
Recent government payroll data has been hugely suspect. On the big release day the numbers are rosy and get reported to much hoopla and "told you" celebration, but then are subsequently revised downwards to little fanfare or notic
749.
by llm_nerd 1y ago
But that's precisely why Meta are the "good guys". They specifically called China the good guys in the same way that Meta is the good guys, though in this case many of the Chinese models are extremely good. Meta has open sour
750.
by llm_nerd 1y ago
The corollary of this is that some people lie and claim that people are paying to shortcut to credibility or legitimacy. And to be clear, I am not talking about this submission (which I didn't read, and from other comments is charity-d

More ›