Imnimo

29 ms·

211.
by Imnimo 2y ago
I don't know whether Sam Altman personally knew about the clause, but clearly OpenAI as an entity knew - it didn't appear out of thin air, and employees who were leaving were clearly away of it. There was even a bit of a minor h
212.
by Imnimo 2y ago
I was and am very skeptical of the "superalignment" plan (which was "build an AI that does alignment research for us, and then ask it what to do"). But it's a bad look for OpenAI if they pledged 20% of their compute
213.
by Imnimo 2y ago
I don't understand how you think AI overviews threaten the existence of developers.
214.
by Imnimo 2y ago
Pluribus is not an LLM and does not operate by next-token prediction. I also don't think it uses human training data - it's trained via self-play by my understanding.
215.
by Imnimo 2y ago
This isn't a fair description of the report. While it has some examples from that sort of games, it has other examples that are not.
216.
by Imnimo 2y ago
I think this says more about the benchmark than the capabilities of the model. If it were the case that 90th percentile performance on the bar exam mean that a model was a 90th percentile lawyer, and we've had these models for 15 month
217.
by Imnimo 2y ago
Yes. To be more specific, it's that nearly all points where the derivative is zero are saddle points rather than minima. Note that some portion of this nice behavior seems to be due to design choices in modern architectures, like resid
218.
by Imnimo 2y ago
In very high dimensional spaces (like trying to optimize a neural network with billions of parameters), to be "in a valley", you must be in a valley with respect to every one of the billions of dimensions. It turns out that in man
219.
by Imnimo 2y ago
Hmm, I do see what you mean.
220.
by Imnimo 2y ago
This sample on Twitter shows how other controllers fail: https://twitter.com/JasonMa2020/status/1786433841613390023 I agree it's hard to tell whether the controller learned with DrEureka would be sufficient w
221.
by Imnimo 2y ago
Another interesting experiment on this front: https://twitter.com/infobeautiful/status/1778059112250589561 One thing I would have liked to see in the blog post is some attention to temperature. It looks like they&
222.
by Imnimo 2y ago
I'd be interested to see how this does on Nicholas Carlini's benchmark: https://nicholas.carlini.com/writing/2024/my-benchmark-for-l... I've tried out some of my own little test prompts, but most of
223.
by Imnimo 2y ago
I honestly don't understand how this is responsive to what I wrote.
224.
by Imnimo 2y ago
I think Zvi is missing some critical points about the bill. For example: >Before initiating the commercial, public, or widespread use of a covered model that is not subject to a positive safety determination, limited duty exemption, a de
225.
by Imnimo 2y ago
I'm extremely skeptical that this shift has anything to do with protest encampments and coddling concerns.
226.
by Imnimo 2y ago
I don't buy a lot of this analysis. >This week, for instance, Google — despite probably being the most progressive or Silicon Valley company, nevertheless fired dozens of employees involved in pro-Palestine/anti-Israel protests
227.
by Imnimo 2y ago
It's not the person who does the fine-tuning I'm worried about, it's the person who releases the base model who the law also makes liable. The point is that, because fine-tuning can trivially induce behavior that satisfies th
228.
by Imnimo 2y ago
I don't agree with that reading. As long as my custom chemical weapon instructions are not publicly available otherwise, then it is surely more difficult to build the weapon without access to the instructions. The line about autonomous
229.
by Imnimo 2y ago
>(2) “Hazardous capability” includes a capability described in paragraph (1) even if the hazardous capability would not manifest but for fine tuning and posttraining modifications performed by third-party experts intending to demonstrate
230.
by Imnimo 2y ago
I was expecting something where the AI chat could run DDG web searches. If it's just a mirror of GPT 3.5/Claude, why do I need DDG to be involved?
231.
by Imnimo 2y ago
One of the challenges here is handling homonyms. If I search in the app for "king", most of the top ten results are "ruler" icons - showing a measuring stick. Rodent returns mostly computer mice, etc. https://
232.
by Imnimo 2y ago
I watched the demo video, and I'm not sure I understand what "seen" means in the title. Is there a camera on this?
233.
by Imnimo 3y ago
I'm really, really struggling to put myself in the frame of mind of a person who feels they need uBlock to block /r/MachineLearning.
234.
by Imnimo 3y ago
Embarrassing for Musk to be so fragile, and embarrassing for Lemon to have been naive enough to think having a show on X would end any other way.
235.
by Imnimo 3y ago
"Trust in AI is down globally from 61 percent in 2019 to just 53 percent, per the Edelman poll." I am having a hard time finding exactly what this number means in the linked report. I see a "50" for AI on page 13 - is th
236.
by Imnimo 3y ago
I don't know if my reading is the intended one, but the way I've always interpreted this is not that random initialization is bad. Rather, the error Sussman makes is saying that a randomly initialized network has no preconceptions
237.
by Imnimo 3y ago
Am I reading right that the network here is an MLP with a single hidden layer of 50 neurons? It's a fun project, but I think the author would have benefitted from spending more time on finding a reasonable network architecture instead
238.
by Imnimo 3y ago
You may be interested in a similar experiment from the Gemini tech report: https://twitter.com/jeffdean/status/1758182184694005787?s=46...
239.
by Imnimo 3y ago
I'm not sure I believe anyone in the field has ever said "AI token". I also don't buy that the term "training data" implies the existence of labeled input/output pairs. Unlabeled data is still training dat
240.
by Imnimo 3y ago
With only $60M a year to go around, I'm not sure my share of comments would even add up to a full dime anyway. They can have it.

More ›