fsndz

18 ms·

211.
by fsndz 2y ago
My point of view: this is a real advancement. I’ve always believed that with the right data allowing the LLM to be trained to imitate reasoning, it’s possible to improve its performance. However, this is still pattern matching, and I suspec
212.
by fsndz 2y ago
because most damage is institutional so most of the work has to be done by institutions (states) and corporations. you can track the carbon footprint of rich people all you want, this won't change a thing until states and corporations
213.
by fsndz 2y ago
This ! And I really like your analogy to computability theory. I wonder if it applies to all economic activities. My experience so far is that verification is nice for simple knowledge tasks. But for the complex ones, doing it from scratch
214.
by fsndz 2y ago
True, the jobs created by AI often seem to prioritize efficiency over meaning, leading to a sense of detachment from human purpose.
215.
by fsndz 2y ago
True. Some people argue that we are quickly moving from a creator economy to a verifier economy. AI generates a bunch of crap and knowledge workers have to verify and edit ( https://www.lycee.ai/blog/the-rise-of-the-veri
216.
by fsndz 2y ago
Great article. A bit in contrast with another report from the US that found AI did not have any impact on recruitment ? ( https://www.lycee.ai/blog/ai-adoption-labor-market-ny-fed-su... ) So what is the key takeaway here
217.
by fsndz 2y ago
Just say you are in founder mode. It's better.
218.
by fsndz 2y ago
I can say that enough. Small Language Models are the future. https://www.lycee.ai/blog/why-small-language-models-are-the-...
219.
by fsndz 2y ago
Jean Piaget said it better: "Intelligence is not what we know, but what we do when we don't know." And what do LLMs do when they don't know, they spit out bullshit. That is why LLMs won't yield to AGI ( https:/
220.
Building an Optimized Question-Answering System with Mipro and DSPY (1)(medium.com)
1 points by fsndz 2y ago|0 comments
221.
by fsndz 2y ago
I don't find pandas intuitive (API simplicity), then you have the hard to debug issues and perf
222.
by fsndz 2y ago
LLMs are great for idea brainstorming. This is a theory I've had in mind for a long time. By simply using tools like ChatGPT, Cohere, Mistral or Anthropic, I quickly realized that language models are quite useful when brainstorming for
223.
by fsndz 2y ago
this explains the success of Groq's ASIC-powered LPUs. LLM inference on Groq Cloud is blazingly fast. Also, the reduction in energy consumption is nice.
224.
by fsndz 2y ago
True. Samsung's Dr. Jung Bae Lee was also talking about that recently. "rapid growth of AI models is being constrained by a growing disparity between compute performance and memory bandwidth. While next-generation models like GPT-
225.
by fsndz 2y ago
exactly what I thought too
226.
by fsndz 2y ago
super nice. why does it degrade quality of image so much, makes it looks obviously AI-generated rapidly.
227.
AI Adoption Hasn't Led to Significant Job Cuts, NY Fed Survey Reveals(lycee.ai)
12 points by fsndz 2y ago|2 comments
228.
by fsndz 2y ago
A recent survey by the Federal Reserve Bank of New York shows that companies adopting artificial intelligence (AI) are not significantly cutting jobs. The findings, based on responses from businesses in the New York-Northern New Jersey regi
229.
by fsndz 2y ago
any resources to self learn biotech and how to use ML in biotech ?
230.
by fsndz 2y ago
At the SEMICON Taiwan CEO Summit, Dr. Jung Bae Lee, Corporate President and Head of Memory Business at Samsung Electronics, stressed that the rapid growth of AI models is being constrained by a growing disparity between compute performance
231.
by fsndz 2y ago
As concerns over artificial intelligence (AI) grow worldwide, both the European Union and California have taken groundbreaking steps to implement comprehensive AI regulation frameworks. These initiatives, which come amid increasing scrutiny
232.
by fsndz 2y ago
exactly, we probably can't even build super intelligence. frankly what we need is more useful tools, we have to quit the idea of creating gods: https://medium.com/@fsndzomga/there-will-be-no-agi-d9be9af44...
233.
by fsndz 2y ago
makes sense if you factor in the cost of renting GPUs to build generative AI models
234.
by fsndz 2y ago
All that money, we are not even sure we can build AGI. What is AGI. Clearly scaling LLMs won't cut it, but VCs keep funding people because they pretend they can build super intelligence. I don't see that happening in the next 5 ye
235.
by fsndz 2y ago
I wish them good luck with it haha, especially given the reliability issues. why would anyone want to 300% given the current reliability challenges ? ( https://www.lycee.ai/blog/ai-reliability-challenge )
236.
by fsndz 2y ago
More like: we don't have AGI yet, and LLMs are not AGI. But what exactly is AGI? Apart from being a useful tool, after considering every possible definition of AGI, it can't be more than human. It can't be superhuman and gene
237.
by fsndz 2y ago
interesting debate around surveillance capitalism, reminded me of this: https://medium.com/discourse/defending-freedom-in-the-era-of...
238.
by fsndz 2y ago
meanwhile police departments are already using that around the world, and they seem conscious about the limitations, and not using it as a 100% accuracy tool. "Assistant Chief of Police of Miami, Armando Aguilar, said in 2023 that Clea
239.
by fsndz 2y ago
even without tech, justice system is already ruining people's lives unjustly. so maybe we have to discuss failure rate, processes in place to correctly use the tech, instead of just saying it has to be bad ? the question is will the te
240.
by fsndz 2y ago
Okay I get it now, GDPR. But is it bad if it is used to catch criminals ? What's a potential misuse of that technology ? Maybe I watched too much of Person Of Internet...

More ›