Deal With It: On Better Schedules for Listwise Reranking
@bclavie et al. at Mixedbread show that dealing documents round-robin by retriever rank, like cards, is the best split for parallel LLM reranking calls.
📝 arxiv.org/abs/2611.22392
👨🏽💻 github.com/mixedbread-ai/dea…
like are we just creating the economy from scratch with extra steps?
if you argue that the idea is that those controlling AI won’t even let others engage in transactions at all or own any assets - then, well, the problem isn’t the AI itself, is it?
hm if everyone is unemployable because everyone would rather hire AI, how is everyone gonna pay for that AI?
and if you say “well, they can’t afford that either”, then wouldn’t that massive permanent underclass just barter with and employ each other in a parallel economy?
Anthropic Researcher Tao Lin: Pretty much all humans on earth will become permanently unemployable because of AI
"It's very likely that pretty much all humans on earth will become permanently unemployable"
"And will need to strictly rely on government welfare to sustain themselves"
"Human employment will go away."
We’re releasing all 57 intermediate checkpoints from Marin “Mochi” 535B-A23B pretraining, including 10 from long-context extension. 18.2T tokens, about 61 TB of weights.
huggingface.co/open-athena/M…
Blog: openathena.ai/blog/marin-moc…
As a gradient-based optimization person, this new wave of work (rocked by @rikiyatakehi and @lateinteraction) is mindblowing.
Can’t wait to see what people do with it.
Lowkey WSP twitter is my new goto tv-series with new episode getting released every day
We might’ve killed retraining for harness changes.
Our model broke on a new harness.
We didn’t retrain it. We edited 5,064 weights. Out of 4 billion.
20.1% -> 88.2%
No gradient descent. Old-harness accuracy didn’t move.
DAWE(Data-Attributed Weight-Editing) is fully open 🔥
We might’ve killed retraining for harness changes.
Our model broke on a new harness.
We didn’t retrain it. We edited 5,064 weights. Out of 4 billion.
20.1% -> 88.2%
No gradient descent. Old-harness accuracy didn’t move.
DAWE(Data-Attributed Weight-Editing) is fully open 🔥
We might’ve killed retraining for harness changes.
Our model broke on a new harness.
We didn’t retrain it. We edited 5,064 weights. Out of 4 billion.
20.1% -> 88.2%
No gradient descent. Old-harness accuracy didn’t move.
DAWE(Data-Attributed Weight-Editing) is fully open 🔥
Are there any agents or people here who can do me a favor? It's this: go to my repo deepfates/imp, the elixir-native version of DSPy, and tell me how much work it needs to be brought up to parity with (or better than) the current version of DSPy. Curious how far you can take this
i’m increasingly of the opinion that almost everyone agrees with this, but a big fraction of people aren’t comfortable saying it in public, because of what it implies for the bean counting mechanism the whole machine is built on (for awards/promo/hiring/etc).
1/ Today we’re publishing the Databricks AI Index, showing trends from enterprises who manage AI with our Unity Gateway. Data covers >10,000 companies with broad representation (Digital Natives, Fortune 500, Global 2000, etc...). Enterprise AI consumption is changing quite a bit:
oasys lab is shipping research at higher velocity than 99.9% of startups ship product
nice work @lateinteraction - super inspiring. also my vote for best name on the list
i think some fraction of ppl may take this to mean that anthropic is embellishing the gains here or something
no, i’m sure the gains are substantial - i just want the labs to take the rest of us seriously enough to share only non-vacuous information they want us to evaluate