Obiwan Santobi (@Osantobi) profile picture

someone has to say what everyone is thinking

Joined May 2021
Photo by Obiwan Santobi (@Osantobi)Photo by Obiwan Santobi (@Osantobi)Photo by Obiwan Santobi (@Osantobi)Photo by Obiwan Santobi (@Osantobi)Photo by Obiwan Santobi (@Osantobi)Photo by Obiwan Santobi (@Osantobi)Photo by Obiwan Santobi (@Osantobi)Photo by Obiwan Santobi (@Osantobi)Photo by Obiwan Santobi (@Osantobi)Photo by Obiwan Santobi (@Osantobi)
Elon is going to find that blue Texas goes much harder than Red Texas
26
These safeguards have turned Claude into a fucking asshole
1
44
Saving this here for later
76
There will come a point when elondre will realize the cursor team is “dookie ass”
29
So basically yes, they are ABSOLUTELY distilling your workflows @ns123abc
This isn't the most notable aspect of today's news, but on the user data issue, there are different kinds of *training on user data* with very different privacy/IP implications. Sadly, AI cos don't like to disclose what they're doing. - pretrain on user data, with users' tokens as prediction targets: high regurgitation risk, improper - use user prompts to distill large models into small ones: low regurg. risk, some companies probably do this - use user traces to construct RL tasks: low regurg. risk, because RL has low memorization abilities, but can extract customer IP, depending on how it's done. Ranges from benign "use explicit user feedback in reward model training" to invasive "upload user's coding environment and commit history to turn into rl envs" "De-identification" is weak -- you can identify someone with a small number of bits, and long traces have more than enough. And it doesn't affect IP leakage concerns.
44
Obiwan Santobi retweeted
This isn't the most notable aspect of today's news, but on the user data issue, there are different kinds of *training on user data* with very different privacy/IP implications. Sadly, AI cos don't like to disclose what they're doing. - pretrain on user data, with users' tokens as prediction targets: high regurgitation risk, improper - use user prompts to distill large models into small ones: low regurg. risk, some companies probably do this - use user traces to construct RL tasks: low regurg. risk, because RL has low memorization abilities, but can extract customer IP, depending on how it's done. Ranges from benign "use explicit user feedback in reward model training" to invasive "upload user's coding environment and commit history to turn into rl envs" "De-identification" is weak -- you can identify someone with a small number of bits, and long traces have more than enough. And it doesn't affect IP leakage concerns.
Two things to distinguish: Did any human or agent look at user data as part of the Navier Stokes effort? No. Do we use user feedback and de-identified data to improve ChatGPT and Codex in a holistic way? Yes. And so does every LLM company.
42
153
34
1,369
285,502
gpt 6 astra is terrible
27
gpt 6 astra is mid
21
Is fable 5.1 really ass cheeks?
24
I had a feeling 💆‍♂️🔮
Dario don’t fumble the bag.. it’s too soon
34
Obiwan Santobi retweeted
Dario don’t fumble the bag.. it’s too soon
1
1
57
Dario don’t fumble the bag.. it’s too soon
1
1
57
Obiwan Santobi retweeted
Google is worth 20T without sundar
1
2
162
Smh
Replying to @thesheetztweetz
They can shake their fist at the sky
1
85
0 to 1 mindset is binary and stupid. Elon mindset of probability non zero unlocks the exponentials which is much better
1
36
Codex fuckin sucks!!!!!
1
57
Codex is not even close to Claude code gpt 5 3 whatever is fucking stupid too. If you hold this line you are an idiot
1
165