wassname Profile Banner
wassname πŸ€”πŸ§ Profile
wassname πŸ€”πŸ§

@wassname

Followers
137
Following
3K
Statuses
1K

Let's align AI better than humans. Transhumanism, curiosity, and the good ending. β†¬βˆžβŠ—Ξ”πŸ¦‹β€³β„΅Ξ¨βŸ²β¬‘πŸ‘οΈπŸ”€πŸ”

onboard GSV Sleeper Service
Joined September 2009
Don't wanna be here? Send us removal request.
@wassname
wassname πŸ€”πŸ§
@wassname
8 hours
@QuintinPope5 @teortaxesTex (with much extinction)
0
0
0
@wassname
wassname πŸ€”πŸ§
@wassname
8 hours
@gm8xx8 1.6B params, so it wont run on mobile
0
0
0
@wassname
wassname πŸ€”πŸ§
@wassname
8 hours
@QuintinPope5 @teortaxesTex tbf geoengineering seems like it might have side effects. releasing lots of iron dust or sulphur are similar to what happened many times in geological history, and so are proven to be effective, but there WERE downsides
2
0
2
@wassname
wassname πŸ€”πŸ§
@wassname
8 hours
RT @QuintinPope5: @teortaxesTex ASI: I've solved climate change. H: No ASI: It's very simple H: No ASI: It's geoengineering + nuclear H: No…
0
3
0
@wassname
wassname πŸ€”πŸ§
@wassname
8 hours
@JohnVial Totally, there are so many algorithmic advanced that haven't been incorporated into the training runs many of the best papers, I find through who replicates them pretty amazing developer
0
0
1
@wassname
wassname πŸ€”πŸ§
@wassname
8 hours
@ziv_ravid @rarefin15 pls gab code
0
0
0
@wassname
wassname πŸ€”πŸ§
@wassname
8 hours
@ziv_ravid Do you apply this to all layers or the last layer?
Tweet media one
0
0
0
@wassname
wassname πŸ€”πŸ§
@wassname
23 hours
RT @davidad: Indeed, in multiple LLM families, many of the neurons toward the very end of each processing step are dedicated to β€œsuppressio…
0
9
0
@wassname
wassname πŸ€”πŸ§
@wassname
23 hours
RT @ziv_ravid: 🧡 I forgot to update, but our paper "SEQ-VCR: Preventing Collapse in Intermediate Transformer Representations" has been acce…
0
23
0
@wassname
wassname πŸ€”πŸ§
@wassname
23 hours
RT @gm8xx8: Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach This model scales test-time computation by reas…
0
52
0
@wassname
wassname πŸ€”πŸ§
@wassname
23 hours
@JohnVial Pretty cool aye. They just reshape the weights into a square, and then say they must be like a blurry convolution of themselves. It's surprising that it doesn't hurt performance!
1
0
2
@wassname
wassname πŸ€”πŸ§
@wassname
23 hours
RT @TechEmails: Mark Zuckerberg messages Facebook engineer April 5, 2012
Tweet media one
0
3K
0
@wassname
wassname πŸ€”πŸ§
@wassname
2 days
RT @wikileaks: USAID (and State) funneled nearly half a billion dollars through this building which is at "876 7th St Arcata, CA 95521-6358…
0
5K
0
@wassname
wassname πŸ€”πŸ§
@wassname
6 days
RT @chelseabfinn: Our first open-source release at Pi πŸ€– - Ο€β‚€ and Ο€β‚€-FAST model weights - code for model, on-robot inference, & fine-tuning…
0
114
0
@wassname
wassname πŸ€”πŸ§
@wassname
6 days
Allen AI is great, this was a great listen
@jxmnop
jack morris
@jxmnop
8 days
even though i do AI research every single day yet, i still feel like i'm constantly behind this was a great listen, especially the first 30 minutes or so where Nathan patiently explains how DeepSeek works. learned a lot i'd listen to something like this every week
0
0
0
@wassname
wassname πŸ€”πŸ§
@wassname
6 days
RT @richten_nach: TL;DR: Chinese LLMs (including DeepSeek) are trained on my illegal archive of books and papers β€” the largest in the world…
0
6
0
@wassname
wassname πŸ€”πŸ§
@wassname
9 days
RT @nrehiew_: So looks like someone can score 93% on the OpenAI Research Engineer interview but not be able to contribute at all to interna…
0
49
0
@wassname
wassname πŸ€”πŸ§
@wassname
10 days
@SchmidhuberAI what's the next thing that you have already done?
0
0
2