coolness
coolness on Hacker News
Super unfortunate. On the other side, I reviewed four papers for a top-tier AI conference and three were clearly fully Claude generated, as in all text, figures, results, everything. Actual good reviewer time is wasted on such papers and your (i hope) human written good paper receives AI responses. It's a sad state of affairs for sure.
Yea but what is the VCT Capabilities Test? According to themselves [1]
> VCT consists of 322 multimodal questions covering fundamental, tacit, and visual knowledge that is essential for practical work in virology laboratories.
So it's just question answering? Do you think that scoring well on this test is equivalent to synthesizing a virus?
[1] https://securebio.org/virologytest/
The title says "hours-long diving" so at least a few hours i guess...
Good work! The animations are super cool, how did you make them?
Slight tangent: i was wondering why DeepSeek would develop something like this. In the linked paper it says
> In production, DeepSeek-OCR can generate training data for LLMs/VLMs at a scale of 200k+ pages per day (a single A100-40G).
That... doesn't sound legal
Great post. Super sad state of affairs but we move on and learn new things. Programming was always a tool and now the tool has changed from something that required skill and understanding to complaining to a neural net. Just have to focus on the problem being solved more.
damn, i was really excited to have a new article by this guy. Makes some of the best articles out there for sure
Wow thats very cool, i was puzzled at first as to why the pokemon types were in Finnish!
Yeah, I don't really understand why someone would make a blog and use AI to write the articles. Isn't having a blog more about the joy of writing and the learning you do while writing it?
Great post and amazing progress in this field! However, I have to wonder if some of these letters were part of the training data for Gemini, since they are well-known and someone has probably already done the painstaking work of transcribing them...
Yeah, even with this fixed it's going to be annoying because it first restarts, so you have to stay at the PC to select windows in grub
Great story and I'm sure the new math chops help her a lot during her studies.
In case anyone else is interested, it is possible to study computer science without any entrance exams due to the Digital Education for All initivative (https://www.helsinki.fi/fi/projektit/digital-education-all in Finnish, sorry). You get the full study right after completing 60 credits (out of a total 180 credits) worth of courses in the first year.
This is a great idea actually, this way you could also enforce that all the words appear _exactly once_ (by rejecting during sampling words that were already in the answer) which seemed to be a significant issue for me when I tried this.
> I expect if you asked it to correct its output in a followup message, it could do so without much difficulty.
I had a similar idea to the author and tried this many times, albeit with the free version of ChatGPT. After getting wrong results, I prompted it to correct them, even telling the model explicitly that a category is wrong or doesn't make sense. Nothing I did made a difference.
My two cents on why this doesn't work has to do with the fact that the answer should contain a discrete set of words given in the prompt, and importantly, they should not be duplicated. I suspect that these currents models are not very good at following the instruction "the token should appear in the answer exactly once"
What an engagingly-written and interesting piece. I really enjoyed the glimpse into the lives of ordinary people somewhere where I may never get the opportunity to travel to.
You can change the language of Google Maps to Finnish, and still keep the phone on English. At least in the latest versions of Android. This makes it pronounce the street names properly
I loved the idea, but on further thought, the AI has a huge advantage in knowing what the engine move is, so they can never lose to incorrectly calling the last move and engine move.
The computer seems to move very slowly, and gets much more time for each move than me. Is this intended, or is it because everyone on HN is trying it out? Might be some sort of client - backend synch issue.
edit: the computer now flagged on my screen, but I did not get any message of a result
Reminds me of the business card scene in American Psycho:
"Look at that subtle off-white coloring. The tasteful thickness of it. Oh, my God. It even has a watermark."
Needs [2020] in the title. Interesting work nevertheless.
The editor itself looks very good - I really like grid-based approach
edit: whoah, the pricing is also incredibly cheap! 18$ a year, most similar services charge tens of dollars a month with similar features
They filter out fast games (bullet and faster) and moves where one has less than 30 seconds total time to make the move. They say this is to avoid capturing "random moves" in time trouble
This, not to mention one could get the GPU usage on the V100 way higher by training with larger batch sizes, which would also make training much faster.
Wow, did you make this using GPT-3? Is the first sentence yours and the rest generated?
This is exactly what most people in this thread are missing. The copyright holder, i.e the creator of the code is never obligated to do anything by the licence he chooses to release with. The licence applied to users of the copyright owners work.