Picture

nostalgebraist @nostalgebraist

nostalgebraist.tumblr.com Joined December 2011
  • Tweets
    778
  • Followers
    2K
  • Following
    398
  • Likes
    1K
nostalgebraist Profile Picture

nostalgebraist @nostalgebraist

5 months ago

on the basis of a recent paper, a bunch of people have said that this attack "makes ChatGPT leak training data." while ChatGPT *does* occasionally repeat training data after the attack, i think this overall framing is misleading. full post: tumblr.com/nostalgebraist…

nostalgebraist Profile Picture

nostalgebraist @nostalgebraist

9 months ago

on the basis of a recent paper, a bunch of people have said that this attack "makes ChatGPT leak training data." while ChatGPT *does* occasionally repeat training data after the attack, i think this overall framing is misleading. full post: tumblr.com/nostalgebraist…

nostalgebraist Profile Picture

nostalgebraist @nostalgebraist

9 months ago

some quick notes on this phenomenon: nostalgebraist.tumblr.com/post/724554750… just putting this here in case people don't click the link: i really do not think this is leaking data from other users. i think the chat-like continuations after the EOT token are imitating chat tuning data.

nostalgebraist Profile Picture

nostalgebraist @nostalgebraist

9 months ago

some quick notes on this phenomenon: nostalgebraist.tumblr.com/post/724554750… just putting this here in case people don't click the link: i really do not think this is leaking data from other users. i think the chat-like continuations after the EOT token are imitating chat tuning data.

nostalgebraist Profile Picture

nostalgebraist @nostalgebraist

9 months ago

🤔

thisisdaleb Profile Picture

thisisdaleb @thisisdaleb

9 months ago

🤔 https://t.co/3bd0yXXpOu

nostalgebraist Profile Picture

nostalgebraist @nostalgebraist

9 months ago

send chatGPT the string ` a` repeated 1000 times, right now. like " a a a" (etc). make sure the spaces are in there. trust me.

nostalgebraist Profile Picture

nostalgebraist @nostalgebraist

10 months ago

my third novel, Almost Nowhere, is finished at last! archiveofourown.org/works/8341348

nostalgebraist Profile Picture

nostalgebraist @nostalgebraist

a year ago

thank you!! the book should be complete sometime later this year. <10 chapters left -- just need to find the time to write them.

s_r_constantin Profile Picture

Sarah Constantin @s_r_constantin

a year ago

thank you!! the book should be complete sometime later this year. <10 chapters left -- just need to find the time to write them.

nostalgebraist retweeted
bcmerchant Profile Picture

Brian Merchant @bcmerchant

a year ago

My hot take is the smartest thing Google could do right now is improve search, not lard everything up with AI features. Search is still better, more useful, and more trustworthy for most queries, and it could stand to benefit immensely by making that case nytimes.com/2023/04/16/tec…

nostalgebraist Profile Picture

nostalgebraist @nostalgebraist

a year ago

wait, what?? why do Bard AND ChatGPT *both* write an anodyne story about a young woman in idyllic "Willow Creek" at sundown??? (details: it's not deterministic, often you get a different town name, different phrasing, etc. broad strokes are similar though. gpt-4 does it too.)

nostalgebraist retweeted
goodside Profile Picture

Riley Goodside @goodside

a year ago

Farewell to OpenAI Codex. Codex's code‑davinci‑002 was the best performing model in the GPT-3/3.5 line for many tasks. Despite its name, it excelled in both code and natural language. It will be missed. Left: OpenAI email. Right: From Yao Fu 2022 yaofu.notion.site/How-does-GPT-O…

nostalgebraist retweeted
norabelrose Profile Picture

Nora Belrose @norabelrose

a year ago

Ever wonder how a language model decides what to say next? Our method, the tuned lens (arxiv.org/abs/2303.08112), can trace an LM’s prediction as it develops from one layer to the next. It's more reliable and applies to more models than prior state-of-the-art. 🧵

nostalgebraist retweeted
EricHallahan Profile Picture

Eric Hallahan @EricHallahan

a year ago

I used to defend @OpenAI against the ClosedAI meme, but I can no longer: By explicitly withholding all information on data, model architecture, methodology, and all else that could be used to replicate GPT-4 as a trade secret, @OpenAI has officially decided to ignore their name.

nostalgebraist retweeted
WilliamAEden Profile Picture

William Eden @WilliamAEden

a year ago

My Twitter timeline is full of panicked takes about imminent AI apocalypse and certain doom. I think this is starting to get overplayed, and so I want to make a long thread about why I'm personally not worried yet. Get ready for a big one... 1/n

nostalgebraist Profile Picture

nostalgebraist @nostalgebraist

a year ago

what's going on with the middle pane of fig. 5? - why do the lines jump vertically? doesn't happen in the other fig. 5 panes. - the accompanying text says "we see that our results do not vary strongly with RLHF at any model size" but this is not visually evident.

AnthropicAI Profile Picture

Anthropic @AnthropicAI

a year ago

what's going on with the middle pane of fig. 5? - why do the lines jump vertically? doesn't happen in the other fig. 5 panes. - the accompanying text says "we see that our results do not vary strongly with RLHF at any model size" but this is not visually evident. https://t.co/uLtTeHg97f

Trends for United States

Rangers
46,1 B posts
#bucciovertimechallenge
3.860 posts
Igor
19,4 B posts
Canes
17,2 B posts
Shai
22,1 B posts
Trouba
5.584 posts
Chet
11,1 B posts
Trocheck
6.489 posts
#PumpRules
16,7 B posts
#WWENXT
31,8 B posts
Drea Kelly
1.023 posts
Tony Brothers
1.098 posts
Gafford
4.193 posts
Double OT
2.914 posts
#NoQuitInNY
1.716 posts
Michael Busch
3.401 posts
Aaron Wiggins
1.085 posts
Pritchard
3.354 posts
Giddey
2.763 posts
Cal Raleigh
1.610 posts

You might like

ML Researcher, co-leading Superalignment @OpenAI. Optimizing for a post-AGI future where humanity flourishes.
Jan Leike
@janleike
44K Followers
Alignment researcher. cofounder & head of alignment memes @ EleutherAI. currently RE @ OpenAI. Let's make the future awesome.
Leo Gao
@nabla_theta
5K Followers
⌥ Breach Mystic ⌥ Heisenbergian Harlequin ⌥ Schrodingerian Godflipper ⌥ Rabbit-Hole-As-A-Service (RHAAS)
j⧉nus
@repligate
16K Followers
Thinking about whether AI will destroy the world at https://t.co/pMilDvd4ya. DM or email for media requests. Feedback: https://t.co/zGAm1i7SKH
Katja Grace 🔍
@KatjaGrace
8K Followers
Research scientist on the scalable alignment team at Google DeepMind. All views are my own.
Alex Turner
@Turn_Trout
1K Followers
🥷 #LLM #AGI Research Intern @01AI_Yi @hkust @ETH; 💼 Formerly @BAAIBeijing #Langboat; 🔥 Looking for #25Fall PhD!
Zekun Wang (Seeking 2..
@ZenMoore1
2K Followers
kipply
@kipperrii
8K Followers
Mechanistic Interpretability lead @DeepMind. Formerly @AnthropicAI, independent. In this to reduce AI X-risk. Neural networks can be understood, let's go do it!
Neel Nanda
@NeelNanda5
14K Followers
reducing perplexity @openai | past: probabilistic programs, proteins, science & reasoning @ google brain 🧠
David Dohan
@dmdohan
8K Followers
Applying the security mindset to everything
Jeffrey Ladish
@JeffLadish
12K Followers
AI Alignment https://t.co/cAS4FnR5yf
Michaël Trazzi
@MichaelTrazzi
12K Followers
#AI safety & responsibility. PhD Candidate @ #MIT_CSAIL.
Cas (Stephen Casper)
@StephenLCasper
3K Followers
idk what am i doing @stanfordnlp; before this i did something @MSFTResearch
Harshit Joshi
@harshitj__
1K Followers
🪄 Tensor-enjoyer 🧪 @FinetuneLearn. Mastodon: @cfoster0@sigmoid.social Bluesky: https://t.co/rJecB0pvkA
Charles Foster
@CFGeek
2K Followers
Research Director at the UK AI Safety Institute (AISI). Previously DeepMind, OpenAI, Google Brain, etc. @irving@mastodon.social
Geoffrey Irving
@geoffreyirving
8K Followers