jsemrau

22 ms·

361.
by jsemrau 3y ago
I guess its a typo https://huggingface.co/codellama/CodeLlama-34b-hf
362.
by jsemrau 3y ago
I guess the main driver here is gaining control / influence ?
363.
by jsemrau 3y ago
DALL-E 3 is using CLIP for synthetic caption generation https://jdsemrau.substack.com/p/paper-review-dall-e-3
364.
by jsemrau 3y ago
Maybe of interest. There has been research by SwissRe on this topic. https://jdsemrau.substack.com/p/paper-review-comparative-saf...
365.
by jsemrau 3y ago
The Internet is on a death-spiral as much as Western Civilisation is on a decline.
366.
by jsemrau 3y ago
Worse are comparison papers where the researchers are evaluating models against a benchmark. In another world, this would have been a blog post.
367.
by jsemrau 3y ago
Swiss Re's opinion is that self-driving cars are safer -> https://arxiv.org/abs/2309.01206 Summary: https://jdsemrau.substack.com/p/paper-review-comparative-saf...
368.
by jsemrau 3y ago
The benefits of using specialized vendors like pinecone is about their offering a combination of performance (clustered/loadbalancing), fast/effective algorithms, and data storage.
369.
by jsemrau 3y ago
Given that I unsuccessfully applied for AIML Annotation roles at Apple for years, I also believe they are building similar tech with internal talent.
370.
Project Gutenberg has implemented one of the worst AI fears of striking actors(qz.com)
43 points by jsemrau 3y ago|74 comments
371.
by jsemrau 3y ago
The Encyclopedia Autonomica defines Agent as a software that is capable of reasoning, using memory, plan its actions, and using tools. Or in short: Agent= Reasoning + Memory + Planning + Tools
372.
by jsemrau 3y ago
My understanding is that the puppeteer records the same movement over and over again and this dataset of trajectories (pose, speed, acceleration) is then "diffused".
373.
by jsemrau 3y ago
I would expect no less from a JP Morgan employee ;-)
374.
by jsemrau 3y ago
I came across this question the other day while discussing AI Agents. Would you chat with an AI agent to be the first line recruiter assuming that you can have a "real" chat with that agent? I.e., the agent ask real questions? OR
375.
by jsemrau 3y ago
The whole notion of "memory" in LLM research solves this problem.
376.
‘Robo-Taxi Takeover’ Hits Speed Bumps(scientificamerican.com)
1 points by jsemrau 3y ago|0 comments
377.
by jsemrau 3y ago
just tried it and its not good. I think one of the better feedback was asking for play an adventure in the style of Space Quest 1
378.
Building a ReAct Agent on LangChain with Memory(jdsemrau.substack.com)
2 points by jsemrau 3y ago|0 comments
379.
by jsemrau 3y ago
I'd think "infrastructur" includes the nice front end and Python API that they have proven to be capable to pull off already.
380.
by jsemrau 3y ago
That should be inversely related to their usage though. I am using HF frequently, but I am not sure if I would pay for it.
381.
by jsemrau 3y ago
I'm on Twitter off/on for 10 years. I still don't get it' value proposition.
382.
by jsemrau 3y ago
Since we are already that far, I am thinking we should by law tag AI generated data in the Metadata.
383.
by jsemrau 3y ago
I see it as an illusion because of the interaction (smile, like), I feel like I have had a social interaction, but in reality I hadn't.
384.
by jsemrau 3y ago
Social Media creates the illusion of social connectivity. While I might tweet a reply to Garry Tan and he might like it. It doesn't mean we are best buddies.
385.
by jsemrau 3y ago
I see China in 2023 where Japan was in 1991. Real Estate is falling apart. But in the end, the are the workbench of the world. They have Billions of people that can potentially work. Some of them are well educated and highly industrious. Ja
386.
by jsemrau 3y ago
I agree. Depth creates much more evergreen content than just "news commentary" or similar.
387.
by jsemrau 3y ago
The problem will be a useful differentiator to other tech blogs.
388.
by jsemrau 3y ago
Expand please. I think the initial hype is gone and we see that most implementations are not that much of a game changer just yet. Stable Diffusion can create images but its still incredibly cumbersome to find the right seed and prompt to
389.
by jsemrau 3y ago
Maybe they were experimenting with reject inference to improve the model but it didn't work?
390.
by jsemrau 3y ago
It might have worked on 2008 levels and then ride the elevator up. But to fuel the growth story they needed to go international where they could not enjoy the same once-in-a-lifetime opportunity.

More ›