abel_

7 ms·

1.
by abel_ 5mo ago
This misses the broader ongoing trend. For a few million dollars, of course you can create a startup that builds tools it can use to more efficiently find code vulnerabilities. And of course you can do this with weaker models with scaffolds
2.
by abel_ 4y ago
On the contrary -- the opposite will happen. There's a decent body of research showing that just by training foundation models on their outputs, you amplify their capabilities. Less common opinion: this is also how you end up with mode
3.
by abel_ 4y ago
This was done so that the IRL robot manipulation tasks could be done fast enough. In the future, we may always need small models mixed with large models for some tasks (e.g., for slow long term planning and fast short term planning), though
4.
by abel_ 4y ago
There's a neat argument against these models doing interpolation: the manifold of the data is so sparse that it's infinitesimally unlikely for a good predictor to be doing interpolation between existing points on the manifold. ht
5.
by abel_ 4y ago
I wonder what the net effect of such pieces of writing is. The problem is that these abstract and contextless statements make sense only if they cause the reader to reflect on some experience, and thus only mildly reinforce currently held b
6.
by abel_ 4y ago
The problem with software-controlled permissions is that nation-state actors (who have unbounded resources) can snoop on your private matters with significantly greater ease. At least with a hardware switch, someone would have to physically
7.
by abel_ 4y ago
While others here have touched on the idea that Codex has changed their coding habits, what I find interesting is that Codex has changed how I write code altogether. For example, I had to connect a database to an API a little while ago. Obv
8.
by abel_ 4y ago
As someone else already mentioned, the scaling laws paint a different story empirically: we haven't hit diminishing returns at all, and there's no end in sight. But more anecdotally, the first applied neural network paper in 1989
9.
by abel_ 4y ago
The most pressing dangers of AI most researchers see: - error rate too high - you can trick a classifier with noise - it's racist sometimes Actual dangers of AI: - stop problem - infeasibility of sandboxing - difficulty of aligning bla
10.
by abel_ 5y ago
While progress has been made in computer vision, that progress has been relatively narrow up until now, and I think the activation energy required to produce this level of quality would be more than it's worth. As others have mentioned
11.
by abel_ 5y ago
Interesting! I hadn't heard this perspective before. Sounds like Goodharting [0] applies to sleep too [0] https://en.wikipedia.org/wiki/Goodhart%27s_law
12.
by abel_ 5y ago
Completely disagree. I agree that the _delivery_ is important, but the interval between when discovering a problem and announcing it should be minimal. There's a huge difference between maintaining the social etiquette of allowing your
13.
by abel_ 5y ago
An interesting reflection is how quickly research around TTS/STT has progressed. I remember reading [0] thinking we were a long ways away. And things will get way better with multi-task learning and multi-modal learning in the coming y
14.
by abel_ 5y ago
I think the distortion of the video-finish camera should be considered when making these estimates. There's quite a lot of it happening in the shot. Narrowing down the specific camera intrinsic parameters may be a challenge. The EXIF d
15.
by abel_ 5y ago
While I see the value in gratitude as described by the author, the lesson learned still rings with a certain amount of complacence with the situation at hand. Had there been more transparency in the bonus program, there's a chance it w