fastball

9 ms·

151.
by fastball 5mo ago
Are they replacing their onboard chips with NVidia chips? AFAICT Tesla has increased investment in their own chip ambitions. Dojo specifically being cut is a massive failure if you are motivated to believe that, but otherwise it looks mor
152.
by fastball 5mo ago
Ok, but Dojo clearly has a spiritual successor in the form of: whatever Tesla is currently using to train FSD models? Like... Tesla kills teams all the time. Sounds like many were absorbed into the rest of the org. Seems like a bit of a not
153.
by fastball 5mo ago
Your previous comment was clearly not constraining itself to the realm of AI chips / FSD. Regardless: How do you measure Dojo, xAI, and Grok being failures? Dojo is training models for FSD, which is in operation today. The chips in Tes
154.
by fastball 5mo ago
Tesla wasn't "successful" before he "took it over" (read: invested most of their seed capital and ran the actual day-to-day operations). SpaceX, founded entirely by him alone, is also the most valuable space technol
155.
by fastball 5mo ago
LLMs mostly only being code-writers was true a year ago, but it is not true now. Now they are tool-callers, which means a coding agent can effectively: run lints/typechecks/tests (and fix resulting errors), dig into observability
156.
by fastball 5mo ago
Yes, and seems pretty clear you can now backpressure the rewrite with tools like miri to have Claude Code automatically improve it.
157.
by fastball 5mo ago
Yes, Claude Code can use a lot of RAM.
158.
by fastball 5mo ago
It's not deterministic, it's probabilistic. Bun does have memory bugs that would not be there if they'd written in safe Rust instead of Zig. You could imagine a scenario where ZigBun has zero memory issues, but it is not th
159.
by fastball 5mo ago
Downstream doesn't imply determinism.
160.
by fastball 5mo ago
How software behaves is very obviously downstream of the tools (in this case programming language) used to build it.
161.
by fastball 5mo ago
Obviously bun having been acquired by Anthropic changes the arithmetic a bit, but I'd love to see the token cost/consumption of this initiative.
162.
by fastball 5mo ago
What are you even talking about?
163.
by fastball 5mo ago
I'm not sure you understand the distinction you are making. The model Google is shipping with Chrome runs on device. ChatGPT does not. The people that dont like data centers should love this feature. Same with people who are concerne
164.
by fastball 5mo ago
ChatGPT (a 3 year old product) has nearly one billion WAU. Some people detest businesses slopping AI at them, but the evidence suggests consumers love using AI, which is presumably one of the primary uses of a micro LLM model that run
165.
by fastball 5mo ago
It is very ironic that this post comes from "The Privacy Guy", given that the whole point of this model is to run inference on your own device rather than sending queries to the cloud, which is also much less power intensive tha
166.
by fastball 5mo ago
It would also imply that it costs Google ~7¢ in only energy cost to deliver that file to you (using average EU energy costs), which is clearly non-sensical given the rates hyperscalers charge for network egress. Additionally, the cited nu
167.
by fastball 5mo ago
> vast majority of laypeople have negative sentiment toward it Citation very much needed. Technologists are not laypeople, and are almost certainly a vocal minority .
168.
by fastball 5mo ago
You could say the same thing about shipping V8 with Chrome. Some users disable JS so shipping V8 with Chrome is additional software they didn't ask for.
169.
by fastball 5mo ago
He is the Chairman of the OpenAI board.
170.
by fastball 5mo ago
How can they not? Surely at GitHub scale there isn't a single component where they were relying on vertical scaling?
171.
by fastball 5mo ago
wat. The protections in place that the OP is talking about are almost entirely due to (not government and company) bad actors.
172.
by fastball 6mo ago
Correct (and secure) code is possible and readily doable. It is unclear if supply chain attacks can ever be fully mitigated.
173.
by fastball 6mo ago
This is what Cloudflare[1] is doing. [1] https://blog.cloudflare.com/post-quantum-roadmap/
174.
by fastball 6mo ago
Surely that is better than burning piles of bodies?
175.
by fastball 6mo ago
A truly unbelievable amount of grift and inefficiency.
176.
by fastball 6mo ago
It's just Simon Willison (the person you are replying to) who always makes a pelican, as his personal flippant benchmark. It's not that deep.
177.
by fastball 6mo ago
What high price? I pay $200/m for an insane number of tokens.
178.
by fastball 6mo ago
Pie-in-the-ski, "humanity needs this so we survive the next 10,000 years" ideas are not good vehicles for fundraising.
179.
by fastball 6mo ago
Being model and provider agnostic are orthogonal concerns. e.g. you can run Claude models on AWS Bedrock giving you provider choice for the same model. Whether or not you need model agnosticism at that point seems like a very different ques
180.
by fastball 6mo ago
https://developer.chrome.com/docs/ai/get-started

More ›