airspresso

7 ms·

61.
by airspresso 1y ago
Messing up with BGP makes you feel alive, that’s for sure. But hey, if you haven’t caused an incident yet, that just means you’re still in onboarding. Those SLA downtime budgets are there to be spent.
62.
by airspresso 1y ago
Yes, for inference the main bottleneck is GPU VRAM and the bandwidth between the GPU cores and VRAM. Ideally you want enough GPU VRAM to be able to load the entire model into VRAM + have room for caching the already-produced output in VRAM
63.
by airspresso 1y ago
The reason is the network effect of enough people having their profiles in one place for genealogy discovery to work as intended. As OP says, he has some relatives in the 23andMe network, but fewer now that people started deleting their dat
64.
by airspresso 1y ago
Many reasons, one being that LLMs are essentially compressing the training data to unbelievably small data volumes (the weights). When doing so, they can only afford to keep the general principles and semantic meaning of the training data.
65.
by airspresso 1y ago
and Claude Code
66.
by airspresso 1y ago
Yes the authors openly acknowledge [0] to be inspired by LSP. [0]: https://www.latent.space/p/mcp
67.
by airspresso 1y ago
I see it much more being a mindset. Either you approach everything with the goal of improving it, or you approach everything with the goal of finding a works-and-is-good-enough solution. One is about maximizing quality and another about min
68.
by airspresso 1y ago
Yes, some companies certainly avoid using Google Cloud due to the strong link between Google and advertising. Especially in Europe you'll find that kind of sentiment. Pushing them to strongly prefer AWS for all things cloud.
69.
by airspresso 1y ago
> In other words, it is expensive, but it is only a money issue. Bingo. That's all that matters.
70.
by airspresso 2y ago
Feedi ( https://github.com/facundoolano/feedi ) Self-hosted. I like the news feed design and deboosting of already-seen entries.
71.
by airspresso 2y ago
This seems misplaced in a discussion about generative AI. Automated systems for handling appels are already a thing and don't necessarily rely on generative AI like LLMs. I agree that bans and appeal handling needs a human in the loop
72.
by airspresso 2y ago
that's value add right there. Summarizing text into audio saves time.
73.
by airspresso 2y ago
Free of cost != free open model. Free of cost means all your requests are logged for Google to use as training data and whatnot. Llama3.2 on the other hand runs locally, no data is ever sent to a 3rd party, so I can freely use it to summari
74.
by airspresso 2y ago
> The difference is that nobody will pay programmers to keep programming once LLMs outperform them. Programmers will simply become as obsolete as horse-drawn carriages, essentially overnight. I don't buy this. A big part of the prog
75.
by airspresso 2y ago
It's good to envision what we'd actually use AGI for. Assuming it's a system you can give an objective to and it'll do whatever it needs to do to meet it, it's basically a super smart agent. So people and companies
76.
by airspresso 2y ago
Agree with your premise, but the value creation math seems off. $0.50/day might become reality for some percentage of US citizens. But not for 3B people around the world. There's also the issue of who gets the benefit of making pe
77.
by airspresso 2y ago
Agree! I remember printing each issue and reading it over and over. So inspiring, convinced me that it's possible to figure out how everything works in tech, down to the wire.
78.
by airspresso 2y ago
lol! Producing academic papers for future training runs then.
79.
by airspresso 2y ago
treasure hunting for those boxes
80.
by airspresso 2y ago
You got it all backwards. Go for a million!
81.
by airspresso 2y ago
Very sceptical of the claims that they don't lose any performance. Sounds like wishful thinking without enough effort put into measuring the performance loss that has to be there due to the heavy quantization. They even dropped computi
82.
by airspresso 2y ago
Doing potentially illegal business verbally with no written trail is a well-known strategy. It is still concealment if those participating do not tell when interrogated. Bottom line is that companies should assume that the government has a
83.
by airspresso 2y ago
Trying the same on https://llama3.replicate.dev/ with Llama 3-70B gives a perfectly fine response with a long poem in Danish. And then it even translates it to English before concluding the response.
84.
by airspresso 3y ago
Nice room for a new player to disrupt then
85.
by airspresso 3y ago
Blog post specifically mentions that using AI to help writing the script does not require labeling the video.
86.
by airspresso 3y ago
Excellent point. With more and more AI-generated content it will be key to be able to tell it apart from the human-generated content.
87.
by airspresso 3y ago
It's not about what is being signed, it's about who signed it and whether you trust that source. I want credible news outlets to start signing their content with a key I can verify as theirs. In that future all unsigned content is
88.
by airspresso 3y ago
No mention of clearly labeling ads made using AI. The deepfake Youtube ads are so annoying. Elon wants to recruit me to his new investment plot? Yeah right.
89.
by airspresso 3y ago
fond memories of Knoppix! Was so useful to boot into that to debug, investigate, fix.
90.
by airspresso 3y ago
This caught my eye as well. I interpreted it as the map only shows countries that have English as primary language. But the article was not precise enough about this. I applaud them for focusing on multilingual performance though, as that i

More ›