Infamous-Bank-7739

Are you employed with your current degree? I would suggest finding projects where the goal is tech oriented, and learn from there.

I think I spent enough hours with Pytorch that I see no reason going to Keras anymore. I really liked it for the very easy documentation, nice blog posts etc. Now I know all that stuff already in Pytorch, so it's hard to see any reason to be excited.

Poking the machine in the right way to produce output that the other machine when poked will output

Well it's going to be one before the other on small projects always. Do you have any good experience on the apple hw then? I can see the benefits of faster memory, but it needs to prove its worth for people to give it any actual attention.

You think people should prefer Mac over more general hw?

Why do the multiplications and sums have a grad? Is this related to how autograd works? I understand it in equation terms, but don't see the necessity in computational terms

I tried running the example on page 2 with chat-gpt 5 times. 5/5 it answered correctly, unlike in the example where they showed chat-gpt giving a bad answer.

Some snippets I found most relevant below, though I think the whole article is worth reading even for the research-oriented people here.

"Although the board included Altman and a close ally, OpenAI President Greg Brockman, it was ultimately controlled by the interests of scientists who worried that the company’s expansion was out of control, maybe even dangerous...

Every time a customer asks OpenAI’s ChatGPT chatbot a question it requires huge amounts of expensive computing power — so much that the company was having trouble keeping up with the explosive demand from users...

From Altman’s point of view, raising more money and finding additional revenue sources were essential. But some members of the board, with ties to the AI-skeptical effective altruism movement, viewed this in tension with the risks posed by advanced AI...

On Friday though, the skeptics won out, and one of the most famous living founders was suddenly relieved of duty...

The source predicted that if the board doesn’t reconsider, a large continent of OpenAI engineers would likely resign in the company days...

A philosophical disagreement wouldn’t normally doom a company that had been in talks to sell shares to investors at an $86 billion valuation, but OpenAI was nothing like a normal company. Altman structured it as a nonprofit, with a for-profit subsidiary that he ran and that had aggressively courted venture capitalists and corporate partners...

Altman needed money, and venture capital firms and big tech companies were interested in backing ambitious AI efforts. To tap that pool of capital, he created a new subsidiary of the nonprofit, which he described as a “capped profit” company. OpenAI’s for-profit arm would raise money from investors, but promised that if its profits reached a certain level — initially 100 times the investment of early backers — anything above that would be donated back to the nonprofit...

The ultimate power at the company rested with the board, which included Altman, Sutskever and president Greg Brockman. The other members were Quora Inc. CEO Adam D’Angelo, tech entrepreneur Tasha McCauley and Helen Toner, director of strategy at Georgetown’s Center for Security and Emerging Technology. McCauley and Toner both had ties to effective altruism nonprofits. Toner had previously worked for Open Philanthropy; McCauley serves on the boards of Effective Ventures and 80,000 Hours...

These efforts, along with the for-profit’s growing success, put Altman at odds with Sutskever, who was becoming more vocal about safety concerns. In July, Sutskever formed a new team within the company focused on reining in “super intelligent” AI systems of the future. Tensions with Altman intensified in October, when, according to a source familiar with the relationship, Altman moved to reduce Sutskever’s role at the company, which rubbed Sutskever the wrong way...

"

Business or politics? With the whole "ClosedAI" debacle in this subreddit and elsewhere, I am smiling to see this being so openly discussed.

Start with BERT as it's easy to pickup. Follow some guide. I agree that you are actually overqualified for this, unless you are afraid of coding.

I bet you can get some midjourney or other to generate those. According to the paper in that ad, they used physics simulations and I doubt you'd want to do that.

The perfect model is one that knows the distribution of the test-set. Can't do this? Well, you can always approximate based on your observations...

Did the open source model license agreement allow to re-use the model without indication of what model was used to finetune?