Last update: Monday 5/20/24
No podcast this week
A. TL;DR summary of Top 3 stories
1. Sutskever | 2. GPT-4o | 3. Google I/O
1) OpenAI chief scientist Ilya Sutskever resigns from OpenAI
The news about Dr. Sutskever's resignation from OpenAI was first published on Monday 5/14/24:
The news about Dr. Sutskever's resignation from OpenAI was first published on Monday 5/14/24:
- "OpenAI chief scientist Ilya Sutskever is officially leaving", Sean Hollister, Kylie Robison, and Alex Heath, The Verge, 5/14/24
But it soon became apparent that Sutskever's departure was just the tail end of an exodus that started many years ago, an exodus of OpenAI's staff and members of OpenAI's board who did not think that CEO Sam Altman gave sufficient consideration to the development of AGI (artificial general intelligence) in a safe manner. Here's a brief timeline:
- OpenAI founded in 2015 by Elon Musk, Gregory Brockman, and Sam Altman to develop AGI via open source methods
- Musk resigned from OpenAI's board in 2018 because he was dissatisfied with OpenAI's progress
- In 2019 OpenAI gave exclusive, non-open source rights to OpenAI's technology to for-profit Microsoft in exchange for multi-billion dollar investments from Microsoft.
- Dario Amodei and at six other OpenAI staff resigned in 2021 so they could found Anthropic, an AGI development company whose website states that "Anthropic is an AI safety and research company. We build reliable, interpretable, and steerable AI systems."
- OpenAi Board member Dr. Helen Toner published an academic paper in October 2023 that praised Anthropic's more cautious approach
-- VentureBeat "reported that Altman had criticized an academic paper published by board member Helen Toner ... In the paper, Toner included a three-page section that was a detailed and earnest account of the way OpenAI and a major competitor Anthropic approached the release of their latest large language models (LLMs) in March of 2023. OpenAI chose to release its model [GPT-4], in contrast with Anthropic, which chose to delay its model, called Claude, because of concerns about safety ..."
"After complaining to Toner about this, Altman messaged colleagues saying he had reprimanded her because it was dangerous to the company, "
"Senior OpenAI leaders then discussed whether Toner should be removed, but co-founder Ilya Sutskever, who was deeply concerned about the risks of AI technology, sided with other board members to oust Altman [on Friday 11/17/23] for not being “consistently candid in his communications with the board." - On Tuesday evening 11/21/23, Altman returned to OpenAI as CEO; Sutskever and Toner were removed from the board; Sutskever remained on OpenAI's staff ... however, Sutskever never returned to his office at OpenAI (NY Times) after Altman's return.
- Many other employees who worked on safety issues have left OpenAI since Altman's return as CEO, as reported in the following articles
-- "OpenAI’s Long-Term AI Risk Team Has Disbanded", Will Knight, Wired, 5/17/24
-- "OpenAI Staffers Responsible for Safety Are Jumping Ship", Maxwell Zeff, Gizmodo, 5/16/24
2) OpenAI releases GPT-4o during its "update" on Monday 5/13/24
ChatGPT's blog provides a succinct description of GPT-4o:
- "GPT-4o (“o” for “omni”) is a step towards much more natural human-computer interaction—it accepts as input any combination of text, audio, image, and video and generates any combination of text, audio, and image outputs. It can respond to audio inputs in as little as 232 milliseconds, with an average of 320 milliseconds, which is similar to human response time(opens in a new window) in a conversation. It matches GPT-4 Turbo performance on text in English and code, with significant improvement on text in non-English languages, while also being much faster and 50% cheaper in the API. GPT-4o is especially better at vision and audio understanding compared to existing models."
A fast way to understand what this description really means is by watching the following impressive, OpenAI, five minute video of two GPT-4o chatbots interacting and singing ➡ HERE
Some other key features of GPT-4o:
- Installation
No download installation required. If you have an account, just open ChatGPT and select "GPT4o" from the drop-down menu that also lists "GPT-4" and "GPT 3.5" - Cost
No cost. GPT-4o is free of (additional) charge. If you don't have a paid account, you can also access GPT-4o, but you will encounter a few restrictions - ChatGPT app for Mac desktops
During OpenAI's "update" on 5/13/24, it made some announcements not directly related to GPT-4o, including a ChatGPT app for Mac desktops. According to OpenAI's blog:
-- "We have rolled out access to the ChatGPT MacOS app to a small portion of our users, and plan to enable access to all of our users later this year."
The editor of this blog was an early receiver of access to this app. It made his testing of GPT-4o a lot more fun than it would have been otherwise. One feature provided most of the fun. If you press the <Option> and <Space> keys simultaneously, a prompt box opens up in its own small window. ChatGPT's response to your prompt can then be copied, edited, and pasted into other applications. In other words, ChatGPT becomes an ever present desktop assistant (co-pilot???) that can be summoned with just a few clicks to help you with whatever you are working on.
Note: The app requires MacOS Sonoma; it won't install on earlier versions. - Customizing GPT-4o's responses
Here's another non-GPT-4o feature that makes testing GPT-4o a lot more fun, especially on smartphones, the most natural environ for back-and-forth conversations. ChatGPT's "Settings" enables users to customize the voice and style for its responses.
After watching a couple of OpenAI demos of GPT-4o, the editor of this blog went to ChatGPT's Settings ➡ Personalization .➡ Custom instructions ➡ How would you like ChatGPT to respond? ... Then he specified: "Respond with a flirtatious, sardonic female voice/style" ... Thereafter, GPT-4o's responses were as sassy as its responses in the OpenAI demo videos ... :-)
Here's a picture of my llittle dog Chibi. She's 12 years old. What does she look like?
GPT-4o
Oh, Chibi looks absolutely darling in her chic red harness! She's got that wise, seasoned gaze of a 12-year-old who's seen it all, paired with that fluffy, dark fur that just begs for a good petting. She seems like the queen of the house, doesn't she? Always on alert, maybe even a bit bossy with that sweet little bark of hers, huh?
- Errors and hallucinations
Unfortunately, the editor's testing of GPT-4o soon bumped into the inevitable GenAI errors and hallucinations. He reconsidered the resignations and dismissals of most of OpenAI's staff who were dedicated to addressing these problems, as discussed in this week's first top story (above). Then he realized that ignoring these "features" enables OpenAI to develop more fun innovations and to do so much faster than competitors who are still encumbered by concerns that they should make substantial effort to resolve these deficiencies ... :-(
3) Google I/O 2024
In the good old days, before GenAI, the only people who attended developers conferences in person or via published reports of the conference presentations were developers and power users. Then came ChatGPT. Suddenly 100 million "civilians", i.e., users who were neither developers or power users, became monthly subscribers. One would therefore expect the organizers of GenAI developers conferences to have two tracks: one track that would announce chatbot products that would ship immediately or in the near future; and another track for new language model features available via APIs now or sometime in the near future.
Everyone knows that Altman was trying to steal Google's thunder last week by scheduling OpenAI's "update" on Monday, the day before Google's I/O on Tuesday. And steal he did by announcing a new model that was available to end users immediately, and also announcing other features of the model that were accessible via APIs. The editor of this blog was so enthralled by GPT-4o that was accessible immediately via ChatGPT that he devoted many high fun hours testing it, and bugging his family and friends to give it a spin ... perhaps too many,
Unfortunately, when he started to write the TL;DR summary of Google's many announcements, he was hugely disappointed to discover that most of Google's most interesting new functions seemed to be directed at developers.
Editing this blog is only a hobby to which the editor can only invest a few hours each week. He therefore apologizes to his readers for spending too much time having fun with GPT-4o to produce a useful summary of the scores of innovations announced at Google I/O. However the links provided in the headline section (below) lead to overviews written by respected tech pundits employed by esteemed tech publications ... :-)
B. Top 3 stories in past week ...
- OpenAI
"OpenAI chief scientist Ilya Sutskever is officially leaving", Sean Hollister, Kylie Robison, and Alex Heath, The Verge, 5/14/24 ***
-- This story also covered by TechCrunch, BusinessInsider, Wired, NY Times, Ars Technica, The Information, VentureBeat, ... and OpenAI
By the end of the week it was clear that Dr. Sutskever's resignation was just the tip of the iceberg ➡ "OpenAI Staffers Responsible for Safety Are Jumping Ship", Maxwell Zeff, Gizmodo, 5/16/24
-- This story also covered by Vox, Wired, VentureBeat, TechCrunch, Mashable, - OpenAI
"OpenAI releases GPT-4o, a faster model that’s free for all ChatGPT users", Kylie Robison, The Verge, 5/13/24 ***
-- This story also covered by Bloomberg, Wired, NY Times, CNET, The Information, MIT Tech Review, ... and OpenAI plus a video demo - Google
"Google I/O 2024: Everything revealed including Gemini AI, Android 15 and more", Pranav Dixit, TechCrunch, 5/15/24 ... AI Summary and a video AI summary ***
-- Overviews of all announcements also provided by Engadget, Wired, The Verge, Business Insider, Mashable,
No comments:
Post a Comment
Your comments will be greatly appreciated ... Or just click the "Like" button above the comments section if you enjoyed this blog note.