alas44
Joined 5 years ago
Karma
399
Submissions
73
Member since
Apr 27
Recent submissions
Comment · 1 month ago
Also cringed at this
Comment · 1 month ago
"AI writing is vague and wrong in hard-to-notice ways"
This one felt to me, at work I lost a lot of time re-reading and correcting AI prose that was launched to "summarize" a collective white paper, and where subtlety and nuance disappeared in ways that harmed the initial ideas/arguments I was writing about, but in sneaky ways.
Comment · 1 month ago
I was downvoted initially, look what OpenAI shared https://openai.com/index/navier-stokes-solution/ ...
"Since August 28 we have been training a new internal model that has exhibited unprecedented performance in our benchmarks, including mathematics. This model’s training is ongoing and its performance continues to improve."
"When a further trained version of our internal model became available over the course of the effort"
"While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models "
Comment · 1 month ago
From OpenAI's post https://openai.com/index/navier-stokes-solution/
"Since August 28 we have been training a new internal model that has exhibited unprecedented performance in our benchmarks, including mathematics. This model’s training is ongoing and its performance continues to improve."
"When a further trained version of our internal model became available over the course of the effort"
"While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models "
Comment · 1 month ago
No, plausible given AI companies want/need session data to train their next models. Probably not someone peeking an eye to sessions directly, but probably not so hard to find the useful sessions in anonymized training data to post train a model on. As stated in the paper, OpenAI did not explicitely denied the researcher sessions were not used for training the model. So either they don't know, or don't want to tell
"I asked whether the model had been trained on, or had access to, our sessions in Codex, into which we had been putting all our drafts for the whole of this project. I was told the model did not look up user data. I asked again, about training, and I did not get an answer."
Let's see what statement OpenAI will come up with for their side of the story
EDIT: precised my thought on user data vs session data
Comment · 6 months ago
Side note, all fusion start ups have built upon decades of science research funded in the ITER program, so opposing ITER to fusion start ups is misleading
Comment · 7 months ago
Not sure how hacker news can effectively protect against what looks like fake users posting LLM generated comments :(
Comment · 7 months ago
Looks like a LLM generated comment
Comment · 9 months ago
There is this project "actually portable executable"/cosmopolitan libc https://github.com/jart/cosmopolitan that allows a compile once execute anywhere style type of C++ binary
Comment · 10 months ago
My brother (a history book worm) writes on geopolitics as a hobby.
He wrote a comprehensive analysis in French on how Greenland will be the tipping point for NATO, the European Union and even Democracy in the USA
I found it very well articulated on the stakes and chain of consequences, so I asked him to translate it in English so I could share it here, thinking it could interest many
Comment · 1 year ago
Echoes the beginning of this science fiction novel! https://en.wikipedia.org/wiki/The_Ice_People_(Barjavel_novel...
Comment · 1 year ago
Severance