5 ms·
Engineers at google have publically stated that the models are too big and are far from their potencial. Glad they're being proven right with every release. Th
by himata4113 5mo ago
Engineers at google have publically stated that the models are too big and are far from their potencial. Glad they're being proven right with every release.
They continue to focus on smaller models while openai and anthropic are increasing compute requirements for their SOTA models.
- stri8ed 5mo agoGiven the cost increase associated with this model, and previous model releases, I think the size is trending upwards, not down.
- himata4113 5mo agoThe speed says otherwise. I think they're increasing costs since they want to start seeing ROI.
- JanSt 5mo agoThose are (mostly) new, faster TPU
- himata4113 5mo agolatest TPU's appear to reach 800tok/s rather than the advertised 300tok/s.
- mgambati 5mo agoThey demoed today 8i running ate 1300 to 1600ish tokens per second. I imagine that is caused by having a single rack serving the model just for the demo.
- himata4113 5mo agoThere's a limit to how much you can "scale" this process, it's linear, but if we did napkin math based on vllm parallel batched streams only lose around ~50% performance compared to single-stream output so doesn't explain the ridicioulusly fast numbers here. I wish google just came out and told us how large their flash model is, because if it's as big or smaller than gpt-5.4-nano that's the real headline here.
- maipen 5mo agoDon’t let that fool yourself. Google will have SOTA models as big as or even bigger than their competitors. They are just refining their current models while they finish training the next generation. They will all come out at about the same time. Anthropic, OpenAi, Google, xAI
- ACCount37 5mo agoAnthropic has been sitting on Mythos for a while now. I guess they don't feel pressured to fuck it ship it until anyone else gets a 10T to work.
- Sevii 5mo agoIt's doubtful they have the compute to make mythos publicly available even after the SpaceX datacenter deal. And why sell it publicly if people are still willing to pay for Opus 4.7?
- outside1234 5mo agoI suspect that Mythos doesn't have a business model that works
- throwa356262 5mo agoAccording to people who have access to Mythos, it is slightly worse than GPT-5.5-xhigh. At least for security tasks. Hold on, I think this claim needs some hard data. Here you go gentlemen: https://www.aisi.gov.uk/blog/our-evaluation-of-openais-gpt-5-5-cyber-capabilities https://www.aisi.gov.uk/blog/our-evaluation-of-openais-gpt-5...
- ACCount37 5mo agoThat claim keeps contradicted hard by other parties, who say Mythos beats 5.5 resoundingly on both autonomous search and discovery and creation of complex exploit chains. There might be a harness difference, but also, this CTF-type benchmark might not capture the capability difference fully.
- howdareme 5mo agoGoogle’s pro models are almost certainly bigger than Openai’s lol
- fikama 5mo agoWhy would that be? I am curious why do you think that.
- ActorNightly 5mo agoBecause TPUs are more efficient, and its cheaper for them to field them in higher quantity since they own the chip.
- mnicky 5mo agoE.g. because they are behind on research and so must compensate with size to achieve similar level of intelligence. At least this is what I heard. For intelligence/size only OpenAI and Anthropic are the frontier. Google has more compute so it can compensate for that with size of the models...
- snovv_crash 5mo agoI'd argue Qwen is pushing the Pareto frontier considerably further when you take size into account.
- Jabbles 5mo ago> Engineers at google have publically stated that the models are too big and are far from their potencial Can you link to a source?
- himata4113 5mo agoI wish I could, it was one of those youtube podcast type interviews with one of the engineers, there was a lot more shared, but that line stuck with me the most.
- Dinux 5mo agoSource please cause i dont believe that for once second
- ActorNightly 5mo agoI mean, yes and no. Nobody really knows the answer to which one is more optimal * Large model trained on a large amount of data across multiple domains, that doesn't need any extra content to answer questions. * Smaller model that is smart enough to go fetch extra relevant content, and then operate on essentially "reformatting" the context into an answer.