After years working on robot memory at @NASAJPL, @GoogleDeepMind , @Meta and @nvidia, @clarkzhango and I kept asking: why does AI still forget?
Today, with @deepakmishraVC, we’re announcing @GraphonAI and our $35M Series A just 6 months after our seed.
• #1 on ATM-Bench, ATM-Bench Hard
• 94% on LoCoMo
• 84% on DolphinBench, 10x cheaper
• A new cost-accuracy frontier across 7 benchmarks
https://t.co/8MkVOtVV0Y
After years working on robot memory at @NASAJPL, @GoogleDeepMind , @Meta and @nvidia, @clarkzhango and I kept asking: why does AI still forget?
Today, with @deepakmishraVC, we’re announcing @GraphonAI and our $35M Series A just 6 months after our seed.
• #1 on ATM-Bench, ATM-Bench Hard
• 94% on LoCoMo
• 84% on DolphinBench, 10x cheaper
• A new cost-accuracy frontier across 7 benchmarks
https://t.co/8MkVOtVV0Y
@garlotic Thanks for digging into the benchmarks! Connecting across formats was the whole point. A lot of what a company knows isn't written down. It's in a recording, a photo, or a video from the field.
@dschewchenko We pull numbers for others from publicly available numbers/systems some of who have hybrid search baseline with re-ranking and some dont (or rather dont disclose)
Internally, we have our own system.
@ThePeterMick BTW, to give you more context,the self supervised graph has many kinds of edges. One of the edges represents time. So when stuff becomes out of data, the graphon simply ignores those edges when traversing the graph