imranq

11 ms·

121.
by imranq 4y ago
Alas my joke didn't really pan out as planned
122.
by imranq 4y ago
was all that work really necessary??
123.
by imranq 4y ago
la li lu le lo
124.
by imranq 4y ago
You can make small LLMs more performant than large ones like GPT-3 by fine-tuning on specific tasks or providing them tools to offload precise calculations: e.g. Toolformer: https://arxiv.org/abs/2302.04761 which uses
125.
by imranq 4y ago
I think we will likely want to know the story that accompanies a painting more than ever. Aesthetics and composition alone will not make a great piece since we can no longer trust that some work was created by a human artist. We know that A
126.
by imranq 4y ago
The internet has about 67% of the world's users, that leaves about 2-3bn not represented. And among those, only about 0.001% actually post and contribute content that is available on the open web, and I'm willing to bet that popul
127.
by imranq 4y ago
This is actually the premise behind Anthropic's Constitutional AI ( https://arxiv.org/abs/2212.08073 ) where they have a set of principles that guide a workflow where the LLM creates the prompts and the critiques. T
128.
by imranq 4y ago
Essentially what large language models allow us to do is move through an n-dimensional space using only words. Somehow we need Google Maps for semantic spaces, but to do that we need to have a strong definition of what a "semantic spac
129.
by imranq 4y ago
I would recommend signing up for teamblind and reaching out to folks on LinkedIn who have the positions / experience you want to get. Look to just learn about their positions and if the rapport is good, get a referral for an applicatio
130.
by imranq 4y ago
Actually this is wrong, after reading the paper: https://arxiv.org/pdf/2302.03027.pdf The main idea is to specify an edit direction which is a broad statement of what you want to do. The model then uses gpt-3 to genera
131.
by imranq 4y ago
I guess this involves training textual inversions which take quite a lot of computation vs other methods like instruct pix2pix which don't need to
132.
by imranq 4y ago
Unfortunately this overhyped launch has started the LLM arms race. Consumers don't seem to care in general about factuality as long as they can get an authoritative sounding answer that is somewhat accurate...at least for now
133.
by imranq 4y ago
Seems like we're at the alchemy stage again with all these blackbox AI models appearing everywhere. Reminds me of this xkcd: https://xkcd.com/1838/ Nit: likely Muslims did not burn down the library of Alexandria,
134.
by imranq 4y ago
Very cool project. I guess LangChain has some code related to this, but I don't think this team published their code anywhere
135.
by imranq 4y ago
SVD is probably the most important theorem in linear algebra. Basically you can take any matrix and find how much it rotates and stretches as a linear transformation
136.
by imranq 4y ago
I predict prompt injection will be a common vulnerability for years to come since the natural language advantages of LLMs is the same advantage hackers will have
137.
by imranq 4y ago
Ah man hope all is well. Layoffs are rough. PM / Dev are two sides of the same coin. One decides what to build, and the other decides how to build it. So instead of picking a role right off the bat, you should pick an industry and go d
138.
by imranq 4y ago
Honestly barring the whole pandemic, 2017 seems like yesterday. The progress on these models has been so fast
139.
by imranq 4y ago
Awesome. does this use only the raw text or does it get help from other summarizers online?
140.
by imranq 4y ago
Second this one, I remember being enthralled in undergrad just going through these lectures. I should give it another watch through since there was a lot I didn't understand the first go around
141.
by imranq 4y ago
This seems like an extremely low yield way to get female attention. A male engineering student would have to take an entire class, study new material, and come up with schemes for seeming like a "chad" for the unlikely chance that
142.
by imranq 4y ago
What is water when we are fish in a dirty river
143.
by imranq 4y ago
I've personally learned a lot from him, plus he seems genuine. Definitely taking a look at this one
144.
by imranq 4y ago
I feel like Google should just divest projects into their own companies with a majority stake in them. That way the high caliber engineers don't have to work on maintaining old software, and the divested company can hire lower paid eng
145.
by imranq 4y ago
I'm guessing privacy issues. After it was revealed how much data TikTok collects even after an Apple review, all apps should be treated with some level of personal scrutiny
146.
by imranq 4y ago
Highly recommend this resource for RL https://spinningup.openai.com/en/latest/
147.
by imranq 4y ago
OpenAI is working on a watermark for their models (if not already there) that would recognize content as AI generated. If its good enough, it should be able to filter out AI content when training new models https://techcrunch.com
148.
by imranq 4y ago
Counter-intuitively, reading Tim Ferris and his DSSS approach made me much better at math Deconstruct: Break down the math you want to know into big problems and concepts. Pick a math-related goal that is Measurable and Time-Bound Selection
149.
by imranq 4y ago
Is there support for Prompt-Tuning? This is also an interesting space to optimize inputs to frozen language models https://ai.googleblog.com/2022/02/guiding-frozen-language-mo...
150.
by imranq 4y ago
Awesome work! I think there should be a version that also refreshes screenshots of DOM elements specifically not just pixels - for example you might want to see how a thumbnail of a video changes over time from a rapidly changing page

More ›