KeplerBoy

7 ms·

151.
by KeplerBoy 6mo ago
Token use cost can easily get as large as dev salaries. Even real businesses care about that.
152.
by KeplerBoy 6mo ago
Yes and the staff will tell them to stop that or charge them extra for it.
153.
by KeplerBoy 6mo ago
There is no workload that latency sensitive is there? If it is you just move the server where it's needed. I.e. you do HFT next to the stock exchange.
154.
by KeplerBoy 6mo ago
Let's revisit that topic if the FED raises interest to 5%.
155.
by KeplerBoy 6mo ago
No you'd not think that. The thought of something not being human written didn't even occur to anyone before decent LLMs came around.
156.
by KeplerBoy 6mo ago
wow, that's weird. One would think that updating the reference table is something a team or individual - who just spent a lot of time and effort on implementing a feature - would also do.
157.
by KeplerBoy 7mo ago
Getting music on an ipod was always a pain unless you bought the music on itunes or ripped a music CD directly with itunes (yes, that was an actual feature. hard to imagine these days). No simple drag and drop onto a mounted USB drive like
158.
by KeplerBoy 7mo ago
Tinygrad sure shared a few opinions on AI PRs on Twitter. I believe the gist was "we have Claude code as well, if that's all you bring don't bother".
159.
by KeplerBoy 7mo ago
For the love of god, don't use these ai generated infographics/diagrams. If that's your bar for quality, I'll think less of your code. I can't help it. Also your saxpy example seems to be daxpy. s and d are short fo
160.
by KeplerBoy 7mo ago
Tried one of those MacBooks Neos in a store a few days ago because I got curious. As a Linux/windows user I was completely baffled that you actually have to click (at least in the default setting) to make a mouse click instead of just
161.
by KeplerBoy 7mo ago
They can, everyone can. Good luck vibe coding marketshare for your new tool.
162.
by KeplerBoy 7mo ago
The dev market? Anthropic's services are arguably more popular among a certain developer demographic. I guess this move might end up in a situation where the uv team comes up with some new agent-first tooling, which works best or only
163.
by KeplerBoy 7mo ago
Most likely because Jetbrains is not for sale. Google almost certainly offered to buy at some point.
164.
by KeplerBoy 7mo ago
mindshare and a central piece of the python package management ecosystem.
165.
by KeplerBoy 7mo ago
ChatGPT had memory for a long time. Claude also had it for quite some time for paying customers.
166.
by KeplerBoy 7mo ago
Pixel buds have this feature, it's called transparency mode. I use it for cycling so I'm still aware of my surroundings in traffic.
167.
by KeplerBoy 7mo ago
Kinda legal insider trading, I guess.
168.
by KeplerBoy 7mo ago
Also heavily used in FPGA based DSP.
169.
by KeplerBoy 7mo ago
I guess because Panther lake ThinkPads won't ship until April, but definitely something to watch out for.
170.
by KeplerBoy 7mo ago
Because AMD had the better chips for the majority of the last decade. Intel has only recently caught up.
171.
by KeplerBoy 7mo ago
OpenAI economics don't really work unless you happen to be OpenAI.
172.
by KeplerBoy 7mo ago
Once all the problems are solved we will be there. Sounds a lot like zeno's paradox. We might be closer than ever but still as far from the goal as ever.
173.
by KeplerBoy 7mo ago
It's safe to assume street view cars capture way more data than the stuff that ends up on the street view product.
174.
by KeplerBoy 7mo ago
No I mean inference. The idea is that inference demand will be massive and a race to the bottom with razor thin margins. Training costs can be amortized over the entire lifetime of the model, but if you lose money on inference or can't
175.
by KeplerBoy 7mo ago
These seems to be much more robotics / autonomous vehicle focused? I don't quite see the mass surveillance angle you get from this you don't already get from cheap ubiquitous cameras, basic computer vision and networking (aka
176.
by KeplerBoy 7mo ago
It might be 400% in the one thing everyone is interested in.
177.
by KeplerBoy 7mo ago
Xeons have a much longer shelf life and diverse workloads. If you order hardware specifically for LLM inference and then some new hardware/model combination is much better at that (which it will be, because a lot of people are working
178.
by KeplerBoy 7mo ago
It all went south when we started to call it "learning" instead of "fitting parameters".
179.
by KeplerBoy 7mo ago
Meh, avx is also just parallelism. That won't get you around Amdahl's law.
180.
by KeplerBoy 7mo ago
What's the most exciting thing going on in TeX right now?

More ›