sherjilozair

10 ms·

1.
by sherjilozair 2y ago
Ace is actually uniquely designed to support that. Our training staff simply record their screens and mouse+keyboard events. We transform that into behavior cloning data to train the model. It's quite easy for us to do custom agents fo
2.
by sherjilozair 2y ago
I'm the founder and CEO of General Agents. Happy to answer questions!
3.
by sherjilozair 3y ago
I’ve been trying to find a maintained typed scripting language. Thank you for building!
4.
by sherjilozair 3y ago
That's not the safety team.
5.
by sherjilozair 3y ago
So this is why my previous company had taken us to a retreat where we tried to brainstorm how to build AGI using post-it notes. We were given 5 minutes and then someone did a crude clustering and picked the biggest cluster.
6.
Some Remarks on Large Language Models(gist.github.com)
182 points by sherjilozair 4y ago|79 comments
7.
by sherjilozair 5y ago
GIL can remain default on. Users can simply disable it for chosen parts of their program.
8.
by sherjilozair 5y ago
You know it’s trivial to add a global lock to any concurrent program, right? What would you lose if other people started writing performant code?
9.
by sherjilozair 5y ago
You don't have to use PICO-8's text editor. You can just open up the p8 file in any text editor and hack away. VS Code has an extension for .p8 files as well which does syntax highlighting correctly.
10.
by sherjilozair 5y ago
Look up iterator, __next__, __iter__.
11.
by sherjilozair 5y ago
DeepMind has already released the open source code and model parameters. The database makes it easier to access the predictions.
12.
by sherjilozair 10y ago
Note that most of recent work in reinforcement learning focus on creating a model which works on all or most of Atari games without using any game-specific knowledge. If researchers were to focus on only Ms. Pac Man, and ignore the "no
13.
by sherjilozair 10y ago
OpenAI ( https://openai.com/ ). OpenAI's research vision is at a similarly ambitious scale as DeepMind. They're also relatively more open in their research, and known to release papers, code, and models quickly. Cur
14.
by sherjilozair 10y ago
You can ride your bike with no handlebars. ;)
15.
by sherjilozair 10y ago
The weights are random. No attempt has been made to make sure the generated image looks aesthetically pleasing. That's the surprising bit: a random neural network generates good-looking images.
16.
by sherjilozair 10y ago
I think Andrej is arguing over and above the reasons you cite. Not only should you learn backprop because of the same reason you learn to do 2+2, but you should learn backprop ALSO because it's a leaky abstraction. This is a non-trivia
17.
by sherjilozair 10y ago
Why doesn't this get talked about more often? That there exists a fundamental VC v/s founder conflict. The VC wants at least one of his startup to go mega, while the founder wants his startup go big enough that his opportunity cos
18.
by sherjilozair 10y ago
It's easier to find fooling perturbations of one image, but not of the whole dataset. I assumed the question was can we use the universal perturbations for robust training? The answer to that is still "no", I think.
19.
by sherjilozair 10y ago
Short answer: No. Computing these perturbations requires an expensive optimization with multiple passes through the dataset, and this would be prohibitively expensive to do in the inner-most loop of training. There are other work in the lit
20.
by sherjilozair 10y ago
bAbI isn't really a "real" problem either, although somewhat better than sorting and the like. bAbI works with extremely restrictive worlds and grammar. In contrast, current speech recognition, language modeling, and object d
21.
by sherjilozair 10y ago
This project (and others in different frameworks) are far from complete. The prize for reproducing WaveNet in open source remains unclaimed.
22.
by sherjilozair 10y ago
The proper YAGNI reply to this would be to use a proper library or work in the right abstraction layer so that you don't have to worry about SQL injection. For instance, Django provides the csrf_token which can be dropped in. That&#x27
23.
by sherjilozair 10y ago
> Googled the product to avoid tracking. You sure you avoided tracking by 'Googling'?
24.
by sherjilozair 10y ago
Gurgaon, India | Softage | Full-stack Developer | Full-time | Internship Softage is India's leading document management company. We have a small tech team which develops our flagship document management workflow software and scales to
25.
by sherjilozair 10y ago
PostgreSQL is the right choice, no matter what.
26.
by sherjilozair 10y ago
+1 to rattray's questions. Some more questions: do you think Django is showing its age? Should new web projects be started in new frameworks like Phoenix, or is Django still good to go?
27.
by sherjilozair 10y ago
What's a better solution than Magento?
28.
by sherjilozair 10y ago
Are you using Meteor for this website?
29.
Why I love basic auth(rdegges.com)
3 points by sherjilozair 10y ago|0 comments
30.
Stop talking about fundraising(inc42.com)
1 points by sherjilozair 10y ago|0 comments

More ›