- 3abitonHonestly fail2ban is amazing. I might doa write up on the countless of attempts on my servers.
- Is there an automated way of doing this?
- Shout outto MIT open courses, and Stanford online for pioneerig this.
- > I get the feeling that it was trained very differently from the other models
It's actually based on a deepseek architecture just bigger size experts if I recall correctly.
- What's the math on the $50k nvidia cluster? My understanding these things cost ~$8k and you can at least get 5 for $40k, that's around half a tb.
That being said, for inference mac still remain the best, and the M5 Ultra will even be a better value with its better PP.
- You could use llama.cpp rpc mode over "network" via usb4/thunderbolt connection
- To be fair it was used a lot during my physics studies. I opted to use it afterwards for integrals and derivations, very powerful.
- Tbh I think theseguidelines are just anticipating future trends.
- > Comparing models, or even different versions of the same model, is a pseudo-scientific mess.
Reminder that in most cases, it's impossible to know if there is cross-contamination from the test set of the public benchmarks, as most LLMs are not truely open-source. We can't replicate them. So arguably it's worse in some cases, pretty much fraud if you account for the VC money pouring in. This is even more evident in unknown models from lesser known institutes like from UAE.
- Now that you mention that, I wonder if LSH would perform better with slightly higher memory footprint
- I read the abstract, while not familiar with the topic, how would we go about limiting the inpact?
- As many pointed out, Macs are decent enough to run them (with maxxed rams). You also have more alternative, like DGX Sparks (if you appreciate the ease of cuda, albeit a tad bit slower token generation performance), or the Strix Halo (good luck with ROCm though, AMD still peddling hype). There is no straitghtforwars "cheap" answer. You either go big (gpu server), or compromise. Either way use either vllm or sglang, or llama.cpp. ollama is just inferior in every way to llama.cpp.
- The hope is lost for Android, there is no moving forward with google antagonizing its foss roots. Libre phone it is. We have to forcibly remove the bandage.
- > Unlike nanochat this is purely vibe-coded, improving vibes by 110%
Karpathy clearly said that it wasn't vibe coded. Apparently it was more time consuming to fix gpt bugs than to do it by himself.
- To be fair finding K is highly domain dependent and I would argue should not be for the analyst (solely) to decide, but with a feedback from domain experts.
- Wait till it gets gobbled by the next gen training data, and embedded in the weights of upcoming LLMs. Paired a clueless vibe coder, with no token limits.
- > Many software engineers now spend as much (or more) time reviewing the output of their own AI tools than their colleagues’ code.
This hits home.
- > Or, to put it another way, OpenAI’s 2025 revenue is on track to only be $3.1 billion more than last year, while its annual operational costs are set to be $24.1 billion more than last year. So, for every dollar of revenue growth OpenAI has, it is costing them $7.77! > > I cannot stress how unprecedentedly dreadful that is. It shows that the promised future investors were piling their money into is a fairy tale. This is a money black hole.
Isn't this a "startup blueprint" for tech companies? Uber, Airbnb, Amazon, etc ... More importantly, AI dominance is more important given the reward?
- > Or, to put it another way, OpenAI’s 2025 revenue is on track to only be $3.1 billion more than last year, while its annual operational costs are set to be $24.1 billion more than last year. So, for every dollar of revenue growth OpenAI has, it is costing them $7.77! > > I cannot stress how unprecedentedly dreadful that is. It shows that the promised future investors were piling their money into is a fairy tale. This is a money black hole.
Isn't this a "startup blueprint" for tech companies? Uber, Airbnb, Amazon, etc ... More importantly, AI dominance is more important given the reward?
- How does this differ from nanochat?