jeswin

20 ms·

421.
by jeswin 3y ago
The only missing thing in NativeAOT (for my use cases) is EF Core. It's getting there. Once EF Core works well on NativeAOT, C# would be a fantastic alternative to Go.
422.
by jeswin 3y ago
> Pretty much as with OSS in general That's not how I feel about OSS - from Operating Systems, to Databases, to Browsers, to IDEs, to tools like Blender etc. Of course there are certain areas where Commercial offerings are better, b
423.
by jeswin 3y ago
> I think the ability of these models to actually use these contexts beyond the standard 8k / 16k tokens is pretty weak. For 32k GPT4 contexts, that's not accurate. GPT4 Turbo is a bit weaker than GPT4-32k, but not to the exten
424.
by jeswin 3y ago
That's not "keeps funding". Which is what GP was objecting to.
425.
by jeswin 3y ago
People have been successfully building on MS APIs for four decades now. I've been for nearly three. What exactly are you saying?
426.
by jeswin 3y ago
As a developer, I'd pay $100 a month for it if they allow 32k+ contexts. At least half my code is now generated from prompts.
427.
by jeswin 3y ago
I would have no problem if a license restricted its commerical use by teracorps, while leaving smaller companies alone. The Llama licence is a step in the right direction.
428.
by jeswin 3y ago
I haven't used Web Components yet, but plan to. If you have worked with or evaluated vanilla Web Components and Lit[1], is there any reason to prefer one of the other? [1]: https://lit.dev/
429.
by jeswin 3y ago
> A stochastic parrot, as described in a seminal paper [1] .... A bit of an exaggeration. The paper reads like a long blog post to me. [1]: https://s10251.pcdn.co/pdf/2021-bender-parrots.pdf
430.
by jeswin 3y ago
Given that AWS still has no billing caps (despite it being one of the most requested features), you're exposing yourself to uncapped downside. In addition to lambdas being a poor architectural choice in most cases, that is.
431.
by jeswin 3y ago
GP asked for the opposite of that. Ligatures are troublesome while writing code.
432.
by jeswin 3y ago
A handgun bullet travels 1200 ft per second. At 1/80000 of a second, the bullet travels 0.18 inches. It won't be very good, but you can actually shoot a bullet with this. With enough light of course.
433.
by jeswin 3y ago
> It helps gpt understand larger code bases by building a "repository map" based on analyzing the abstract syntax tree of all the code in the repo. What I do with codespin[1] (another AI code gen tool) is to give a file/fi
434.
by jeswin 3y ago
For code generation, just as exciting as the context length is the new cut off date (2023-04? wow!). It knows about new APIs, frameworks, techniques etc.
435.
by jeswin 3y ago
> It also has reified generics (IMHO a good thing from a performance perspective), which bakes in a particular way of doing parametric polymorphism. Reified generics need additional support, and if you're using type erasure you didn
436.
by jeswin 3y ago
> Sure, if you define "a bit" to be %30 then you can say that. 30% is roughly the difference to be expected between N5 -> N5P/N4 -> N3. That was the original argument - "TSMC". About laptop battery life etc
437.
by jeswin 3y ago
1) M2 uses N5P. N4 has little advantage over N5P. 2) 7950HS is 35-54w. You'll see several results above/around 14k.
438.
by jeswin 3y ago
> A laptop that can match the best of the competition's desktop is pretty big advantage IMHO. 7945HX3D (a 5nm 55W laptop part) scores 15-16k on GB6. M3 Max 3nm is just a little better. Almost all of which can be attributed to TSMC 3
439.
by jeswin 3y ago
Since we agree on the non-linear relationship between TDP and benchmark scores, can we also agree that further reduction in power (to say 45w) would only cause a small drop in performance? For argument's sake, let's triple the 10%
440.
by jeswin 3y ago
> True, but then again you get 400GB/sec of bandwidth For what, other than toy AI inference? More HEDT users would go with more RAM at DDR5 speeds than 400GB/sec of bandwidth. As you can see from Geekbench, the additional band
441.
by jeswin 3y ago
> The M3 Max scores higher on GB6 than the 7950X, full stop: https://browser.geekbench.com/processors/amd-ryzen-9-7950x As you can see in this list [1], there are several systems which exceed 21k. The 19k scores are
442.
by jeswin 3y ago
So a bit faster than 7950x in eco mode (which has a much lower tdp), and the 7950x process is two generations old. It's clear that Apple's advantage is mostly just one thing: "TSMC".
443.
by jeswin 3y ago
> Is that really a fair comparison, though? GP is highlighting a problem with Apple, and it's quite true. I have a licenses to Macomedia Fireworks (from 2010) for macOS and Windows. Windows version works still flawlessly. macOS drop
444.
by jeswin 3y ago
CredCore | UI/UX Senior Developer | Remote | Anywhere | Full-Time CredCore is a well-funded early-stage startup based out of New York City building solutions for the $10 trillion US debt sector. Our product suite helps funds and compan
445.
by jeswin 3y ago
Given the number of times it just fails with large prompts on 32k contexts, I'm not sure if they'ready for this. In my experience, if you're consuming 20k+ tokens failure rate is more than 50%.
446.
by jeswin 3y ago
That AI is transformative for development is not in doubt any more. Just this past week, I've been able to build two medium sized services (a couple of thousand lines of code in python, a language I hadn't used for more than a dec
447.
by jeswin 3y ago
> This principle applies at the end of a war as well. Because it would have entailed the awful human costs of an invasion, Truman’s demand for Japan’s unconditional surrender to end the war was indefensible. When 75 million people have d
448.
by jeswin 3y ago
> You gave them a massive lead with a novel approach... From everything I've read/watched, the original researchers didn't know at that point their work was going to be so impactful. In hindsight, yeah. But otherwise no.
449.
by jeswin 3y ago
> It seems unlikely that any significant number of developers quit using SO in favor of blindly trusting ChatGPT. Most people I know have done exactly that.
450.
by jeswin 3y ago
Thanks. :) Happy to help if you need any help - my email is in my profile.

More ›