The bitter lesson is finally coming for the self-driving cars. The vision stack, 3D maps, lane selection grammar, occupancy networks, it’s maybe all about to give way to a single GPT looking at camera feeds and predicting the next steering wheel adjustment.
It’s mostly a latency problem at this point. The models are too big to run locally, but given that open-weight models like Qwen already exist, an open-weight, low latency equivalent to Astra can’t be too far out.
As somebody working near the field, I do enjoy the fun of dreaming bespoke vision and autonomy algorithms (if I didn’t, I wouldn’t work in the field to begin with!). But I would drop it all in a heartbeat for a robot that works well. Robust, resilient robots would be such an incredible advance that the ‘how’ doesn’t matter. All of the nonsense from the current AI hype cycle would be worth it if it cashed out in Robots That Actually Work.
How do you justify all these data centers when models are becoming good enough to run locally and other models are being directly "burned" into chips (model on chip)? Open weight models are almost as good as the closed ones now and they are free/uncensored.
The only thing DCs will still be need for is training, everything else will be done locally on your own hardware.
Open models are still behind February's Mythos.
Whether they can narrow that gap in the future, or OpenAI and Anthropic widen the gap with access to more compute and their better models assisting in the research process, remains to be seen.
At this time I see no reason to believe these data centers won't be in high demand.
Extremely unlikely seeing what the Chinese have been able to do with the limited resources they have. The creativity in finding improvement such as what deepseek has released is incredible. At this point it's a bet on the looser if you think the open models won't catch up and surpass the closed ones.
valine · · focus · HN ↗
It’s mostly a latency problem at this point. The models are too big to run locally, but given that open-weight models like Qwen already exist, an open-weight, low latency equivalent to Astra can’t be too far out.
sebastos · · focus · HN ↗
user43928 · · focus · HN ↗
Not to mention construction, infrastructure, agriculture, manufacturing, logistics...
--
AI hype cycle? It's working today.
It's optimizing ML model graphs for me while I type this, and it already cut inference time from 30s to 18s.
--
Some people act like there was no way for the AI labs to make back the $800B being invested in data center construction this year.
If we look at global GDP, it's $126T, and even a 5% productivity gain would correspond to $6T.
Is that impossible? Is it guaranteed to all crash? I don't think so.
sschueller · · focus · HN ↗
The only thing DCs will still be need for is training, everything else will be done locally on your own hardware.
This bubble will burst and it will be ugly.
user43928 · · focus · HN ↗
Whether they can narrow that gap in the future, or OpenAI and Anthropic widen the gap with access to more compute and their better models assisting in the research process, remains to be seen.
At this time I see no reason to believe these data centers won't be in high demand.
sschueller · · focus · HN ↗
Extremely unlikely seeing what the Chinese have been able to do with the limited resources they have. The creativity in finding improvement such as what deepseek has released is incredible. At this point it's a bet on the looser if you think the open models won't catch up and surpass the closed ones.
user43928 · · focus · HN ↗
Rumored breakthroughs in efficiency were reported a few times.