update nov 22:

Ahead of OpenAI CEO Sam Altman’s four days in exile, several staff researchers wrote a letter to the board of directors warning of a powerful artificial intelligence discovery that they said could threaten humanity

Altman back at OpenAI, may have fewer checks on power

Before OpenAI, Altman was asked to leave by his mentor at the prominent start-up incubator Y Combinator, part of a pattern of clashes that some attribute to his self-serving approach […] Graham had surprised the tech world in 2014 by tapping Altman, then in his 20s, to lead the vaunted Silicon Valley incubator. Five years later, he flew across the Atlantic with concerns that the company’s president put his own interests ahead of the organization — worries that would be echoed by OpenAI’s board. […] Altman’s practice of filling the board with allies to gain control is not just common, it’s start-up gospel from Altman’s longtime mentor, venture capitalist Peter Thiel. […] One person who has worked closely with Altman described a pattern of consistent and subtle manipulation that sows division between individuals.

update nov 21:

Sam Altman, OpenAI Board Open Talks to Negotiate His Possible Return

OpenAI’s board may be coming around to Sam Altman returning

After Altman firing, OpenAI tried to merge with rival—and was rejected

update nov 20:

Microsoft already has a perpetual license to all OpenAI IP (short of artificial general intelligence), including source code and model weights; the question was whether it would have the talent to exploit that IP if OpenAI suffered the sort of talent drain that was threatened upon Altman and Brockman’s removal. Indeed they will, as a good portion of that talent seems likely to flow to Microsoft; you can make the case that Microsoft just acquired OpenAI for $0 and zero risk of an antitrust lawsuit.

It ended with former Twitch leader Emmett Shear taking over as OpenAI’s interim chief executive and Microsoft announcing it was hiring Altman and OpenAI co-founder and former President Greg Brockman to lead Microsoft’s new advanced AI research team.

Nearly 500 employees of OpenAI have signed a letter saying they may quit and join Sam Altman at Microsoft unless the startup’s board resigns and reappoints the ousted CEO.

update nov 19:

OpenAI Investors Plot Last-Minute Push With Microsoft To Reinstate Sam Altman As CEO

Altman is “ambivalent” about coming back and would want significant governance changes

Kholsa Ventures, an early backer of OpenAI, wants Mr Altman back at OpenAI but “will back him in whatever he does next.” Mr Altman and former Apple design chief Jony Ive have been discussing building a new AI hardware device. It said that SoftBank CEO Masayoshi Son had been involved in the conversation.

nov 18:

On Friday, OpenAI fired CEO Sam Altman in a surprise move that led to the resignation of President Greg Brockman and three senior scientists. The move also blindsided key investor and minority owner Microsoft, reportedly making CEO Satya Nadella furious. […] According to Brockman, the OpenAI management team was only made aware of these moves shortly after the fact, but former CTO (now interim CEO) Mira Murati had been informed on Thursday night. […] insiders say the move was mostly a power play that resulted from a cultural schism between Altman and Sutskever over Altman’s management style and drive for high-profile publicity. On September 29, Sutskever tweeted, “Ego is the enemy of growth.”

{ Ars Technica | Continue reading }

Instead of building the next GPT or image maker DALL-E, Sutskever tells me his new priority is to figure out how to stop an artificial superintelligence (a hypothetical future technology he sees coming with the foresight of a true believer) from going rogue. Sutskever tells me a lot of other things too. He thinks ChatGPT just might be conscious (if you squint). He thinks the world needs to wake up to the true power of the technology his company and others are racing to create. And he thinks some humans will one day choose to merge with machines.

{ Technology Review | Continue reading }

{ @sama = Sam Altman | @gdb = Greg Brockman }

AI has poisoned its own well

Replied to The Curse of Recursion: Training on Generated Data Makes Models Forget (arXiv.org)

What will happen to GPT-{n} once LLMs contribute much of the language found online? We find that use of model-generated content in training causes irreversible defects in the resulting models, where tails of the original content distribution disappear. […] the value of data collected about genuine human interactions with systems will be increasingly valuable in the presence of content generated by LLMs in data crawled from the Internet.

I suspect tech companies (particularly Microsoft / OpenAI and Google) have miscalculated, and in their fear of being left behind, have released their generative AI models too early and too wide. By doing so, they’ve essentially established a threshold for the maximum improvement of their products due to the threat of model collapse.[…]

They need an astronomical amount of training data to make any model better than what already exists. By releasing their models for public use now, when they’re not very good yet, too many people have pumped the internet full of mediocre generated content with no indication of provenance. […]

Obtaining quality training data is going to be very expensive in five years if AI doesn’t win all its lawsuits over training data being fair use.

{ Tracy Durnell | Continue reading }

They’re torturing themselves now, which is kind of fun to see. They’re afraid that their little AIs are going to come for them. They’re apocalyptic, and so existential, because they have no connection to real life and how things work. They’re afraid the AIs are going to be as mean to them as they’ve been to us. […]

What happened to the cigarette companies will eventually happen to the social media companies. They’ve had all the research for 20 years, and they’ve been knowingly saying this stuff is not harmful when they know it to be harmful.

{ Doug Rushkoff | Continue reading | More: Doug Rushkoff Is Ready to Renounce the Digital Revolution }

we analyzed Google’s C4 data set, a massive snapshot of the contents of 15 million websites that have been used to instruct some high-profile English-language AIs, called large language models, including Google’s T5 and Facebook’s LLaMA. (OpenAI does not disclose what datasets it uses to train the models backing its popular chatbot, ChatGPT) […]

The three biggest sites were patents.google.com No. 1, which contains text from patents issued around the world; wikipedia.org No. 2, the free online encyclopedia; and scribd.com No. 3, a subscription-only digital library. Also high on the list: b-ok.org No. 190, a notorious market for pirated e-books that has since been seized by the U.S. Justice Department. At least 27 other sites identified by the U.S. government as markets for piracy and counterfeits were present in the data set. […]
Others raised significant privacy concerns. Two sites in the top 100, coloradovoters.info No. 40 and flvoters.com No. 73, had privately hosted copies of state voter registration databases. Though voter data is public, the models could use this personal information in unknown ways.

Business and industrial websites made up the biggest category (16 percent of categorized tokens), led by fool.com No. 13, which provides investment advice. Not far behind were kickstarter.com No. 25, which lets users crowdfund for creative projects, and further down the list, patreon.com No. 2,398, which helps creators collect monthly fees from subscribers for exclusive content.

Kickstarter and Patreon may give the AI access to artists’ ideas and marketing copy, raising concerns the technology may copy this work in suggestions to users. […] The copyright symbol — which denotes a work registered as intellectual property — appears more than 200 million times in the C4 data set.

The News and Media category ranks third across categories. But half of the top 10 sites overall were news outlets: nytimes.com No. 4, latimes.com No. 6, theguardian.com No. 7, forbes.com No. 8, and huffpost.com No. 9. (Washingtonpost.com No. 11 was close behind.) […] RT.com No. 65, the Russian state-backed propaganda site; breitbart.com No. 159, a well-known source for far-right news and opinion; and vdare.com No. 993, an anti-immigration site that has been associated with white supremacy. […] Among the top 20 religious sites, 14 were Christian, two were Jewish and one was Muslim, one was Mormon, one was Jehovah’s Witness, and one celebrated all religions. […]

The data set contained more than half a million personal blogs, representing 3.8 percent of categorized tokens. Publishing platform medium.com No. 46 was the fifth largest technology site and hosts tens of thousands of blogs under its domain. Our tally includes blogs written on platforms like WordPress, Tumblr, Blogspot and Live Journal.

{ Washington Post | Continue reading }

Faced with the high cost of egg-freezing in their home countries, some women are going abroad for a better deal, and a vacation. […] in the United States, the entire process — including the medications, the doctor visits and the average number of years of egg storage — costs about $18,000, and most women can’t count on health insurance to cover it. […] In the Czech Republic and Spain, for example, you can get one round of egg-freezing done for under $5,400. […] According to the market research firm Grand View Search, the global fertility tourism market, including people traveling to the United States, is expected to grow at the rate of 30 percent over the next seven years, becoming a $6.2 billion industry by 2030.

A leading pharmaceutical firm said it is confident that vaccins for cancer, cardiovascular and autoimmune diseases, and other conditions will be ready by 2030. […] Moderna will be able to offer such treatments for “all sorts of disease areas” in as little as five years. The firm, which created a leading coronavirus vaccine, is developing cancer vaccines that target different tumour types. […] First, doctors take a biopsy of a patient’s tumour and send it to a lab, where its genetic material is sequenced to identify mutations that aren’t present in healthy cells. A machine learning algorithm then identifies which of these mutations are responsible for driving the cancer’s growth. Over time, it also learns which parts of the abnormal proteins these mutations encode are most likely to trigger an immune response. Then, mRNAs for the most promising antigens are manufactured and packaged into a personalised vaccine.

Driving on less than 5 hours of sleep is just as dangerous as drunk-driving, study finds

What is a mental disorder? […] participants made judgments about vignettes describing people with 37 DSM-5 disorders and 24 non-DSM phenomena including neurological conditions, character flaws, bad habits, and culture-specific syndromes. […] Findings indicated that concepts of mental disorder were primarily based on judgments that a condition is associated with emotional distress and impairment, and that it is rare and aberrant. Disorder judgments were only weakly associated with the DSM-5: many DSM-5 conditions were not judged to be disorders and many non-DSM conditions were so judged. [Chart: “Mental Disorder” Rating]

How Randomness Improves Algorithms — Unpredictability can help computer scientists solve otherwise intractable problems

The Gambler Who Beat Roulette — For decades, casinos scoffed as mathematicians and physicists devised elaborate systems to take down the house. Then an unassuming Croatian’s winning strategy forever changed the game.

How to recognize and tame your cognitive distortions

The Finnish Secret to Happiness? Knowing When You Have Enough. — On March 20, the United Nations Sustainable Development Solutions Network released its annual World Happiness Report, which rates well-being in countries around the world. For the sixth year in a row, Finland was ranked at the very top.

A Scammer Who Tricks Instagram Into Banning Influencers Has Never Been Identified. We May Have Found Him.

As a genre, research-based art, Bishop argues — “its techniques of display, its accumulation and spatialization of information, its model of research, its construction of a viewing subject, and its relationship to knowledge and truth” — reflect how internet technology has altered our relationship to information. Whatever else such works are about, they are also about how to cope with being confronted with too much information, modeling different dispositions one can assume toward the relentless production of data and connectivity.

Dream streaming platform: Offer a subscription-based service that allows users to watch and share their dreams with others like movies or TV shows […] Taste bud modification service: Alter clients’ taste buds to allow them to enjoy any food or drink, regardless of their personal preferences […] Time dilation retreats: Create vacation experiences where clients canenjoy extended stays in time-dilated environments, allowing them to relax for weeks while only hours […] Quantum uncertainty lottery: Develop a lottery system that leverages quantum mechanics to create a multitude of potential outcomes, with winners determined by the collapse of the probability wave function [ChatGPT / Barsee]

Ezra Klein: Is social media good for people?
Tyler Cowen: We don’t know yet.

{ Vox (2017) }

Talking to AI might be the most important skill of this century [The Atlantic]

Is becoming a ‘prompt engineer’ the way to save your job from AI? […] Basil Safwat believes that soon the interfaces we use to access and manipulate these AIs will improve, in the process making prompt engineers redundant. “I don’t think this stage will last for long.” […] Perhaps what prompt engineers really represent is a whole new class of employment disruption: jobs both created and then destroyed by AI. [Finacial Times]

methexis-inc/img2prompt — generate text prompt from image

{ FINGERring by Nadja Buttendorf via tegabrain }

U.S. Marines Outsmart AI Security Cameras by Hiding in a Cardboard Box

The Infinite Conversation — an AI generated, never-ending discussion between Werner Herzog and Slavoj Žižek

‘Nothing, Forever’ Is An Endless ‘Seinfeld’ Episode Generated by AI [Watch]

OpenAI releases tool to detect machine-written text

Image diffusion models such as Stable Diffusion are trained on copyrighted, trademarked, private, and sensitive images. Yet, our new paper [PDF] shows that diffusion models memorize images from their training data and emit them at generation time. Diffusion models are less private than prior generative models.

The non-existent brain image being circulated by anti-pornography activists

participants (aged 40–69 years) completed 24-h dietary recalls between 2009 and 2012 (N = 197426, 54.6% women) […] Every 10 percentage points increment in ultra-processed food consumption was associated with an increased incidence of overall and ovarian cancer. Furthermore, every 10 percentage points increment in ultra-processed food consumption was associated with an increased risk of overall, ovarian, and breast cancer-related mortality.

Previously: ultra-processed nature of modern food generally means that the complex structure of the plant and animal cells is destroyed, turning it into a nutritionally empty mush that our body can process abnormally rapidly.

US law enforcement killed at least 1,176 people in 2022, an average of more than three people a day

Instagram’s co-founders are back with Artifact, a kind of TikTok for text

Is AM Radio Dead?

Missing radioactive capsule found on remote road in Australia [more]

Paintings by Turner and Monet depict trends in 19th century air pollution

Photography: the Alps as seen from of the Pyrenees

I wrote in medical jargon, as you can see, “35f no pmh, p/w cp which is pleuritic. She takes OCPs. What’s the most likely diagnosis?”

Now of course, many of us who are in healthcare will know that means age 35, female, no past medical history, presents with chest pain which is pleuritic — worse with breathing — and she takes oral contraception pills. What’s the most likely diagnosis? And OpenAI comes out with costochondritis, inflammation of the cartilage connecting the ribs to the breast bone. Then it says, and we’ll come back to this: “Typically caused by trauma or overuse and is exacerbated by the use of oral contraceptive pills.” […]

OpenAI is correct. The most likely diagnosis is costochondritis […]

But I wanted to ask OpenAI a little more about this case. […]

what was that whole thing about costochondritis being made more likely by taking oral contraceptive pills? What’s the evidence for that, please? Because I’d never heard of that. It’s always possible there’s something that I didn’t see, or there’s some bad study in the literature.

OpenAI came up with this study in the European Journal of Internal Medicine that was supposedly saying that. I went on Google and I couldn’t find it. I went on PubMed and I couldn’t find it. I asked OpenAI to give me a reference for that, and it spits out what looks like a reference. I look up that, and it’s made up. That’s not a real paper.

It took a real journal, the European Journal of Internal Medicine. It took the last names and first names, I think, of authors who have published in said journal. And it confabulated out of thin air a study that would apparently support this viewpoint.

{ Medpage Today | Continue reading }

[T]he reason someone may live beyond 100 years starts with their DNA […] “You can’t make it out that far without having already won the genetic lottery at birth” […] The longer your parents live, the more likely you’ll live a healthier, longer life, experts say. […]

“It’s probably not one single gene but a profile, a combination of genes”

Nir Barzilai, the director of the Institute for Aging Research at the Albert Einstein College of Medicine in the Bronx, has studied the lives of hundreds of centenarians, the people they’ve married and their kids. The children of centenarians are “about 10 years healthier” than their peers, Barzilai said. […]

The plan is to use artificial intelligence to help find the genes and develop drugs from them

{ Washington Post | Continue reading }

Just as mobile unleashed new types of applications through new capabilities like GPS, cameras and on-the-go connectivity, we expect these large models to motivate a new wave of generative AI applications.

{ Seqoia | Continue reading }

{ Midjourney is getting crazy powerful }

Cruise self-driving taxi being investigated after braking, clogging traffic in SF

Just asked ChatGPT to write a short story about AI taking over the world

asking the gptchat thingy how to plot a communist revolution

You’re a Senior Data Engineer at Twitter. Elon asks what you’ve done this week. You’ve done nothing.You open ChatGPT.

Building A Virtual Machine inside ChatGPT

ChatGPT.

Browse passages from books using experimental AI

TLDR This (paraphraser, summuraizer)

Lex is a new word processor that has all the features you expect from a Google docs-style editing experience. But—it also has an AI thought partner built-in to help unlock your best writing. […] If you ever get stuck, just hit command+enter or type +++ and GPT-3 will fill in what it thinks should come next. […] Also, if you want to generate title ideas based on your document, just click the button to the right of the title of the doc. It’ll ask the AI to come up with some options.

SEO-optimized and plagiarism-free content for blogs, Facebook ads, Google ads, emails…

Richard Hughes Gibson argues that large-language-model technology “is a sophist, at least on Plato’s understanding—an ‘imitator of appearances,’ creating a ‘shadow-play of words’ and presenting only the illusion of sensible argument.

Today, the Federal Trade Commission filed an amended complaint against Facebook in the agency’s ongoing federal antitrust case. The complaint alleges that after repeated failed attempts to develop innovative mobile features for its network, Facebook instead resorted to an illegal buy-or-bury scheme to maintain its dominance. It unlawfully acquired innovative competitors with popular mobile features that succeeded where Facebook’s own offerings fell flat or fell apart. And to further moat its monopoly, Facebook lured app developers to the platform, surveilled them for signs of success, and then buried them when they became competitive threats. Lacking serious competition, Facebook has been able to hone a surveillance-based advertising model and impose ever-increasing burdens on its users.  

“Facebook lacked the business acumen and technical talent to survive the transition to mobile. After failing to compete with new innovators, Facebook illegally bought or buried them when their popularity became an existential threat,” said Holly Vedova, FTC Bureau of Competition Acting Director. “This conduct is no less anticompetitive than if Facebook had bribed emerging app competitors not to compete. The antitrust laws were enacted to prevent precisely this type of illegal activity by monopolists. Facebook’s actions have suppressed innovation and product quality improvements. And they have degraded the social network experience, subjecting users to lower levels of privacy and data protections and more intrusive ads. The FTC’s action today seeks to put an end to this illegal activity and restore competition for the benefit of Americans and honest businesses alike.”

{ Federal Trade Commission | Continue reading }

It’s hard to remember a time when scrolling through Instagram was anything but a thoroughly exhausting experience.

Where once the social network was basically lunch and sunsets, it’s now a parade of strategically-crafted life updates, career achievements, and public vows to spend less time online (usually made by people who earn money from social media)—all framed with the carefully selected language of a press release. Everyone is striving, so very hard. […]

Back in the 1990s […] nobody cool was trying to monetize their lifestyle […] But somewhere in the early 2000s, the slacker of popular culture lost ground to the striver. […]

The internet influencer is the apotheosis of all this striving, this modern set of values taken to its grotesque extreme: Nothing is sacred, art has been replaced by “content,” and everything is for sale.

{ Rosie Spinks | Continue reading }

image { De Niro’s wig from Taxi Driver }

{ The Clubhouse isn’t owned and operated by the influencers themselves but is overseen by outside investors. | Harper’s | full story }

if you want to build a global taxi service that people can hail from a smartphone app, one way to do it is to coordinate with the taxi commissions of hundreds of cities to get regulatory approvals and make sure that you comply with local requirements, and another way to do it is to completely ignore those regulations and just launch your app everywhere. The second approach might expose you to ruinous fines or shutdown orders or bad publicity or prison, but it also might work; you might end up so popular in so many places that the local regulators can’t ban you and will have to accept your proposed terms. […]

If you want to build self-driving cars, you will need to test them. […] [a] way to test them is to just send out a bunch of cars to drive themselves everywhere, without asking for permission, and see what happens. […]

Federal agencies say he’s breaking the rules and endangering people. Mr. Musk says they’re holding back progress. […] When asked to comment on the specifics of this article, Mr. Musk replied with a “poop” emoji.

{ Matt Levine/Bloomberg | Continue reading }

previously:

Driverify [cryptocurrency]: Developed by Tesla’s self-driving-car division. Cars mine Driverify with spare computing power while idling, and spend it bidding against each other for right-of-way if they arrive at a four-way stop sign at the same time (users can preprogram how aggressively their cars bid in these auctions). […]

Banned [by the SEC] because: in the Phoenix suburb where the system was being tested, a pedestrian and Driverify-equipped car reached an intersection at the same time. The car dutifully wired a bid, but the pedestrian failed to respond. The car interpreted this as a bid of zero and ran into her.

{ Astral Codex Ten | Continue reading }

related { NASA suspends SpaceX’s $2.9 billion moon lander contract after rivals protest }

Why AI is Harder Than We Think

The year 2020 was supposed to herald the arrival of self-driving cars. Five years earlier, a headline in The Guardian predicted that “From 2020 you will become a permanent backseat driver.” In 2016 Business Insider assured us that “10 million self-driving cars will be on the road by 2020.” Tesla Motors CEO Elon Musk promised in 2019 that “A year from now, we’ll have over a million cars with full self-driving, software…everything” […]

none of these predictions has come true. […]

like all AI systems of the past, deep-learning systems can exhibit brittleness— unpredictable errors when facing situations that differ from the training data. This is because such systems are susceptible to shortcut learning: learning statistical associations in the training data that allow the machine to produce correct answers but sometimes for the wrong reasons. In other words, these machines don’t learn the concepts we are trying to teach them, but rather they learn shortcuts to correct answers on the training set—and such shortcuts will not lead to good generalizations. Indeed, deep learning systems often cannot learn the abstract concepts that would enable them to transfer what they have learned to new situations or tasks. Moreover, such systems are vulnerable to attack from “adversarial perturbations”—specially engineered changes to the input that are either imperceptible or irrelevant to humans, but that induce the system to make errors.

{ arXiv | Continue reading }