With a webcam, you usually have a fixed background, which is an easier problem to solve. Here is a project from 2020 that can do 4k at 30fps https://github.com/PeterL1n/BackgroundMattingV2
The syntax was specifically chosen to avoid confusion with other tag formats. XML could show up in any user or LLM-generated text and would mess up the message format. Some variations even used unicode symbols like "｜"…
Context is just one long string. You can do whatever you want with it. Training helps to conform to specific syntactic idiosyncrasies, but LLMs are large enough these days that they can follow pretty much any…
Amusingly, the agent also spoke German with an English accent. I guess it takes a bit to switch contexts.
I'll believe in AGI when Google can figure out how to leverage their model to make their user experience not suck.
> LLMs are super lazy about caring about wall clock time. My bandaid for problem is to prefix every bash command with "timeout 60 " at the harness level and remind the LLM to hurry up. Of course, this is not a perfect…
What <thing> are we talking about? The only number in the same ballpark that I'm aware of is 2.9B commits per month (not per day): https://github.blog/news-insights/company-news/the-august-17... The number of PRs sits…
llama.cpp is already extremely fast for single-token responses (<5 ms). I can't see Jev being faster when taking network latency into account, except maybe for multimodal inputs.
> I was training custom ML models back in 2017. Maybe you are a good person to ask my question then. I have not looked into Jev much, but is it much different from using a regular LLM and constraining its token output…
> Deepseek Flash plans were very cheap when they first launched. They raised prices That does not necessarily mean that they are losing money. Almost every company was compute-constrained at some point. > Your link is…
> Deepseek also came with heavily subsidized plans, at least at first. Do you have a source for that? I got a source straight from the hoses mouth, which claims that DeepSeek-R1 was highly profitable: > If all tokens…
LLMs can decompile assembly code quite well these days, so I doubt that WebAssembly will be a hurdle.
I could find are a bunch of Photopea repositories on GitHub, but the authors are all either from China or Russia, so getting damages for infringement will be difficult to enforce. Hiring a lawyer sounds like a waste of…
Only 17%? I'd say driving down the price of AI helps at least 95% of the world's population. It certainly helps me.
I was wondering whether the model could help me with German bureaucracy. Unfortunately, the answer is "no". More specifically, I asked the model about what I should put on my contact page, which can cost you in the…
This article suggests a massive oversupply of GPUs for inference, but I don't see any mention of training. I was under the impression that training makes up a significant fraction of OpenAI's/Anthropic's compute budget.…
How did you measure energy usage?
Called it three weeks ago: https://news.ycombinator.com/item?id=49595431
The full-response APIs of many providers have been inadequate for a while now because their timeout intervals do not account for lots of thinking. You can use the streaming API to avoid timeouts.
Those RTX 5090 numbers are bad. You can get over 200 tps with ninfer using NVFP4 and MTP.
I don't know the tool used here, but DeepSeek-V4.1-Flash with bash can create a somewhat similar SVG: https://asdf10.com/diagram.svg Disclaimer: I cut it off after a few minutes because I got impatient. It could have…
I agree that ASCII is definitely cooler, for example when personalizing the hashed data as in your example.
> 1024 bit keys with recordings of not too old data can probably be found I think GitHub might turn into a scary vector of supply chain attacks in the foreseeable future. There is a five digit number of users still…
Here's a larger partial hash collision (108 trailing bits): echo 23ca73454a1b981fe51cad0dbd05f4e696795ba67abb28c61aea1a024e5bbeca | xxd -r -p | sha256sum echo…
When running this ssh command in qterminal, I can not click on the URL because it refreshes faster than I can right-click and click on "Open Link". Do I have to type the URL by hand or is there a workaround?
With a webcam, you usually have a fixed background, which is an easier problem to solve. Here is a project from 2020 that can do 4k at 30fps https://github.com/PeterL1n/BackgroundMattingV2
The syntax was specifically chosen to avoid confusion with other tag formats. XML could show up in any user or LLM-generated text and would mess up the message format. Some variations even used unicode symbols like "｜"…
Context is just one long string. You can do whatever you want with it. Training helps to conform to specific syntactic idiosyncrasies, but LLMs are large enough these days that they can follow pretty much any…
Amusingly, the agent also spoke German with an English accent. I guess it takes a bit to switch contexts.
I'll believe in AGI when Google can figure out how to leverage their model to make their user experience not suck.
> LLMs are super lazy about caring about wall clock time. My bandaid for problem is to prefix every bash command with "timeout 60 " at the harness level and remind the LLM to hurry up. Of course, this is not a perfect…
What <thing> are we talking about? The only number in the same ballpark that I'm aware of is 2.9B commits per month (not per day): https://github.blog/news-insights/company-news/the-august-17... The number of PRs sits…
llama.cpp is already extremely fast for single-token responses (<5 ms). I can't see Jev being faster when taking network latency into account, except maybe for multimodal inputs.
> I was training custom ML models back in 2017. Maybe you are a good person to ask my question then. I have not looked into Jev much, but is it much different from using a regular LLM and constraining its token output…
> Deepseek Flash plans were very cheap when they first launched. They raised prices That does not necessarily mean that they are losing money. Almost every company was compute-constrained at some point. > Your link is…
> Deepseek also came with heavily subsidized plans, at least at first. Do you have a source for that? I got a source straight from the hoses mouth, which claims that DeepSeek-R1 was highly profitable: > If all tokens…
LLMs can decompile assembly code quite well these days, so I doubt that WebAssembly will be a hurdle.
I could find are a bunch of Photopea repositories on GitHub, but the authors are all either from China or Russia, so getting damages for infringement will be difficult to enforce. Hiring a lawyer sounds like a waste of…
Only 17%? I'd say driving down the price of AI helps at least 95% of the world's population. It certainly helps me.
I was wondering whether the model could help me with German bureaucracy. Unfortunately, the answer is "no". More specifically, I asked the model about what I should put on my contact page, which can cost you in the…
This article suggests a massive oversupply of GPUs for inference, but I don't see any mention of training. I was under the impression that training makes up a significant fraction of OpenAI's/Anthropic's compute budget.…
How did you measure energy usage?
Called it three weeks ago: https://news.ycombinator.com/item?id=49595431
The full-response APIs of many providers have been inadequate for a while now because their timeout intervals do not account for lots of thinking. You can use the streaming API to avoid timeouts.
Those RTX 5090 numbers are bad. You can get over 200 tps with ninfer using NVFP4 and MTP.
I don't know the tool used here, but DeepSeek-V4.1-Flash with bash can create a somewhat similar SVG: https://asdf10.com/diagram.svg Disclaimer: I cut it off after a few minutes because I got impatient. It could have…
I agree that ASCII is definitely cooler, for example when personalizing the hashed data as in your example.
> 1024 bit keys with recordings of not too old data can probably be found I think GitHub might turn into a scary vector of supply chain attacks in the foreseeable future. There is a five digit number of users still…
Here's a larger partial hash collision (108 trailing bits): echo 23ca73454a1b981fe51cad0dbd05f4e696795ba67abb28c61aea1a024e5bbeca | xxd -r -p | sha256sum echo…
When running this ssh command in qterminal, I can not click on the URL because it refreshes faster than I can right-click and click on "Open Link". Do I have to type the URL by hand or is there a workaround?