causal

50 ms·

931.
by causal 2y ago
I appreciate that you're taking feedback seriously, and it sounds like you're making some good changes. But frankly, all my goodwill was burnt up in the days I spent trying to make LangChain work, and the number of posts I've
932.
by causal 2y ago
Perhaps my biggest criticisms of 1984 is that it tends to make the reader fixate on some group of evil conspirators at the top, at the expense of understanding the banality of evil. I think this portrayal sows distrust in real democracy thr
933.
by causal 2y ago
Yeah I thought the consensus against LangChain was formed a year ago, surprised to still be seeing these articles.
934.
by causal 2y ago
Appears to be a one shot PoC, don't see it on the author's Github, though @jonathanyc does appear to be active on HN still.
935.
by causal 2y ago
Ironically, with growth triggered by a Microsoft investment
936.
by causal 2y ago
Furthermore, the AWS estimates are also really poorly done. Using EKS this way is really inefficient, and a better comparison would be AWS Bedrock Haiku which averages $0.75/M tokens: https://aws.amazon.com/bedrock&#x2F
937.
by causal 2y ago
You're right. In fact, using EKS at all is silly when AWS offers their Bedrock service with Claude Haiku (rated #19 on Chat Arena vs. ChatGPT3.5-Turbo at #37) for a much lower cost of $0.75/M tokens (averaging input and output lik
938.
by causal 2y ago
No way you need $3,800 to run an 8B model. 3090 and a basic rig is enough. That being said, the difference between OpenAI and AWS cost ($1 vs $17) is huge. Is OpenAI just operating at a massive loss? Edit: Turns out AWS is actually cheaper
939.
by causal 2y ago
Came here to say this. No way you need to spend more than $1500 to run L3 8B at FP16. And you can get near-identical performance at Q8 for even less. I'm guessing actual break-even time is less than half that, so maybe 2 years.
940.
by causal 2y ago
Relevant Schneier post https://www.schneier.com/blog/archives/2023/12/ai-and-mass-s...
941.
by causal 2y ago
Not only is it wrong, it's backwards: the US has some of the only refineries capable of processing heavier crudes, and so other countries depend on us to refine it, especially Canada. US is a net energy exporter, and ship out a lot of
942.
by causal 2y ago
Refusing to answer any question would result in a perfect score for the first chart since it says nothing of specificity
943.
by causal 2y ago
Designers try? It's not like engineers can always prevent bad management from interfering with good product
944.
by causal 2y ago
Apple really likes to limit hardware interoperability because hardware is where they make their $, but there's a lot more overlap in software compatibility between OSX and Linux than Windows and anything else. Just having the familiar
945.
by causal 2y ago
What do you mean about NVIDIA? I find their drivers have become pretty good. Especially so if you're using them with containers.
946.
by causal 2y ago
I'm glad that it's shining a light on the reality of Windows 11 as a subscription and data collecting vehicle. If you still hate Windows least, that's almost certainly because it's what you know best. I work with Windows
947.
by causal 2y ago
It's a human-preference benchmark. Useful but not sufficient.
948.
by causal 2y ago
Qwen 2 uses GQA, has a much longer context length, and covers many new languages. You can't do that with the same architecture and dataset.
949.
by causal 2y ago
Maybe but I don't see how this release would support that conclusion- their 72B model surpasses Llama 3 70B on so many metrics by such a wide margin I find it a little hard to believe.
950.
by causal 2y ago
Even if that were true (it's not; arch and dataset differ considerably from model to model) the release of new foundation models has a lot of other important factors to consider (e.g. performance, license, size).
951.
by causal 2y ago
The way SSR has normalized intermixing of front- and back-end development has always bothered me for reasons I haven't been able to articulate well, but this is a good one.
952.
by causal 2y ago
Exactly. Windows is almost a loss leader at this point.
953.
by causal 2y ago
Agree on the Apple strengths, disagree that MS is scared. PCs just aren't where they make most of their money anymore. Enterprise is extremely lucrative for them, and their headstart in the AI space may be just what they need to close
954.
by causal 2y ago
Why not a Macbook? I get why people might hate Apple as much as MS, but it's certainly more privacy friendly than Windows and more user friendly than Linux. And yes I know Linux has come a long way - I use it daily - but there is still
955.
by causal 2y ago
Completely agree, but for that reason also think we need to be more specific when discussing AI safety. Both "AI" and "safety" are overbroad in the extreme.
956.
by causal 2y ago
I appreciate that this also avoids other big tech solutions; i.e. MS and Apple.
957.
by causal 2y ago
Hmm yeah these particular shadows don't look that well aligned to me, but the hypothesis certainly warrants more searching
958.
by causal 2y ago
Not sure I see it? Maybe would help to have someone highlight specific lines/shadows. I can maybe see it if I let my brain see some alignment and ignore the mostly not-aligned bits.
959.
by causal 2y ago
Yeah, private H&R Block and Intuit have built empires on upselling their products to taxpayers.
960.
by causal 2y ago
Yeah, premature DRY is a pet peeve of mine. Especially since the "size" of the code necessary to trigger DRY is totally subjective: some people apply DRY when they see similar blocks of code, others are so averse to repetition the

More ›