Contact Me By Email

Saturday, April 15, 2023

One year later, Elon Musk’s Twitter is losing support - The Washington Post

A year ago, Musk asked, ‘Is Twitter dying?’ He may have his answer.

Twitter has been dramatically transformed under Musk and few — even among some in the billionaire’s corner — say the changes have been for the better

Updated April 15, 2023 at 1:32 p.m. EDT|Published April 15, 2023 at 8:00 a.m. EDT

A collage of the twitter icon, Elon Musk, and a circular timeline with various buttons and icons from Twitter's interface
(Emma Kumer/The Washington Post; Jim Watson/AFP/Getty Images; Amy Osborne for The Washington Post; iStock)

"SAN FRANCISCO — A year ago this month, Elon Musk turned to his favorite social media site with a question. “Is Twitter dying?” he asked.

On the anniversary of his hostile takeover bid to buy the social media company, he may have his answer.

Twitter has been dramatically transformed under Musk and few — even amongsome in the billionaire’s corner — say the changes have been for the better. In recent weeks, government agencies, news organizations and powerful social media influencers have questioned the usefulness of the platform, with some major players publicly abandoning their accounts or telling users they can’t rely on it for urgent information.

Advertisers have fled in droves over Musk’s policy changes and erratic behavior on the site, causing advertising revenue to recently drop by as much as 75 percent, according to a person familiar with the matter who spoke on the condition of anonymity to share sensitive internal information. Rounds of layoffs have left Twitter operating with a skeleton staff of 1,500 — an 80 percent reduction — and so riddled with bugs and glitches that the site goes down for hours at a time. Meanwhile, the company’s valuation has cratered, Musk has said, to less than half the $44 billion he paid when he bought the company roughly six months ago.

The resulting changes have left a key venue for seeking critical information unreliable, hindering the flow of critical information.

“I’d say the pain level of Twitter has been extremely high,” Musk said in an interview this week with the BBC, assessing his first six months in charge. “It’s been really quite a stressful situation.” But he added that advertisers were returning and that he anticipates a roughly “break even” financial picture, adding: “Overall, I think the trend is very good.”

Even some Musk fans see things quite differently. Musk has garnered a reputation as a business genius with a Midas touch, but his erratic decision-making at Twitter has taken some of the sheen off his supposed brilliance.

“I am disappointed that he seems to have made as many mistakes as he’s made with Twitter. I don’t think the product’s gotten materially better,” said one person in Musk’s orbit who initially cheered the takeover, speaking on the condition of anonymity to offer a candid assessment of his tenure.

“He overpaid. He got a lot of bad press,” the person added. “It’s surprising.”

Musk and Twitter did not respond to a request for comment.

Saturday morning, Musk tweeted that he’d deleted his list of blocked accounts on Twitter. “Negative feedback received on this platform is great for reducing ego-based errors,” he added.

Even Musk’s purchase of the site was steeped in drama. It began a little over a year ago, when it was revealed that Musk, then the world’s richest person, had purchased a more than 9 percent stake in Twitter, making him its largest individual shareholder. Days later, he launched a hostile takeover bid, valuing the company at $54.20 per share. Months of back and forth ensued — including an attempt by Musk to back out — before the deal ultimately closed at the end of October.

As he launched his bid a year ago, Musk wrote that Twitter’s failure to adhere to free speech — likely a reference to its banning problematic accounts — “fundamentally undermines democracy” given the site’s role in fostering public debate. He accused Twitter’s leaders of tolerating armies of phony accounts, known as bots, and pledged to “defeat the spam bots or die trying!” Under his leadership, he said, Twitter would seek to “authenticate all real humans.”

Roughly six months in, Musk has kept some of those promises. He has restored many previously banned accounts, including that of former president Donald Trump. And his subscription model aims to tackle bot accounts, building off attempts his staff has made to target geographic centers of spam. Next week, he says Twitter will switch largely to a paid subscription model for verifying the authenticity of accounts.

But some of those decisions are noticeably shifting the conversation on the site and carrying unintended consequences such as outages.

Former employees who spoke on the condition of anonymity for fear of retribution previously said that Musk had gutted many of the company’s core functions, like risk assessment on new product launches and deleting some user data in line with federal requirements. Twitter internally is chaotic, they said, with littlecoherent strategy other than the imperative to respond to Musk’s latest whim.

“The magic is gone and the carnival barker’s in charge,” one of the former employees said. It “feels like it’s gone from the center of all public conversation to some regrettable networking event where everyone’s got business cards but no one’s got a job.”

Recently, some users have noticed bizarre glitches. Some found their timelines were populated with users they did not follow and more recently, some noticed an inability to reply to tweets.

Earlier this week, users of Twitter Circle — a feature that limits content shared on Twitter to a smaller groups — found their content was instead being shared across the site, including explicit content. Days later, Twitter users widely reported a new problem: They could no longer reply to tweets in their timelines using a web browser.

Part of the chaos stems from Musk’s decision to shift from a chronological feed to a new “For You” feed that relies on an algorithm to surface content. It fundamentally changes the way the site functions, inundating users with a flood of tweets from people they don’t follow — or necessarily want to see. It relies on signals users send when they interact with other content on the site. Even the old chronological “Following” feed is populated with recommendations for tweets from outside users, a Washington Post analysis found.

A previous Post analysis found Twitter was amplifying hate speech in its “For You” timeline, thanks in part to Musk’s decision to restore thousands of previously suspended accounts.

Under Musk, Twitter expanded its overall emphasis on recommendations and added a twist. Code released in late March includes a “multiplier” for tweets from users who pay to subscribe to Twitter Blue, attaining the blue check mark that was previously limited to verified users.

Those users’ tweets would be set to a default value of two, apparently pushing them into the feeds of users who don’t follow them at a likelihood twice as high as that of people who don’t pay. It’s not clear what value, if any, is in use today.

Musk’s changes to the verification model — as well his decisions driving away some news organizations — are making the site less reliable as a destination for information, some experts and users say.

This week, both NPR and PBS said they would stop tweeting from their main accounts after Twitter labeled them “government-funded media.” Twitter first labeled NPR as “state-affiliated,” a label generally reserved for government-run propaganda outlets like Russia’s RT and the Chinese Communist Party’s People Daily, before changing the label when NPR pushed back.

“We are not putting our journalism on platforms that have demonstrated an interest in undermining our credibility and the public’s understanding of our editorial independence,” NPR said in a statement.

PBS spokesman Jason Phelps said it stopped tweeting when it learned of the government-funded label and has “no plans to resume at this time.”

Musk took aim at NPR’s stance in a tweet, pointing out that NPR’s website calls federal funding “essential” to public radio’s service.

“What have you got against the truth @NPR?” he wrote.

A different change is threatening Twitter’s usefulness to the National Weather Service, which has long used automatic tweets to communicate urgent news about extreme weather to the public. Twitter users have had access to systems that allow them to push automated posts by connecting to external sources of information so that the Weather Service and meteorologists can send out quick posts when tornadoes or floods hit.

As weather conditions change, the agency relies on these automated posts to keep people up to date, sometimes sending dozens a day.

Twitter said last month it would limit automated tweets to 1,500 a month, and charge $100 per month for anyone who wanted to send up to 50,000. After the change takes effect, the Weather Service said, its automated tweets about severe weather “may not be posted.” Officials said Twitter told them no exceptions would be made to the new limits.

Twitter’s tweaks are already causing smaller issues for weather watchers: James Spann, chief meteorologist for ABC 33/40 in Birmingham, Ala., said Twitter integration with Slack is broken, a problem for him because he uses the chat tool to communicate with weather spotters on the ground. They would use the auto tweets to push storm warnings to the spotters.

“That’s really aggravating,” he said. “One day, we had a tornado event here and we had no warnings on the Slack channel.”

On Friday, the U.S. National Tsunami Warning Center tweeted it would also be affected by the automated tweet limit, adding “we will make every effort to continue manual posts.” San Francisco Bay Area transit system BART also tweeted Friday it would stop using its account for sharing alerts, although it later walked back its position saying some of its access to the site was restored.

Twitter unveiled another change this month that riled up some power users. After the newsletter site Substack unveiled a feature that appeared to resemble a Twitter clone, Twitter started limiting engagement on tweets that included Substack links. Twitter’s apparent hostility toward Substack prompted the journalist Matt Taibbi — who Musk selected to write portions of the “Twitter Files” — to announce he was remaining at Substack and shifting to its Twitter clone.

Twitter is also aggressively trying to monetize, to the frustration of some users. Twitter recently announced that it would limit access to its application programming interface (API), software tools that allow outside researchers and developers to collect and analyze data, and charge a fee in most cases.

One researcher said Twitter told them enterprise access will now cost an exorbitant price that will limit the ability of researchers to study the spread of online conversations, misinformation and harms.

“As a research institute, I don’t have $42,000 a month to pay them for less data than we get now,” said Caitlin Watkins, the executive director of the Observatory on Social Media at Indiana University, which studies media and technology in society and makes tools to analyze social media. That could force them to shut down tools built for journalists and researchers to access data, as well as hurt their efforts to research topics like the spread of misinformation.

“I don’t think anyone is overselling it or being dramatic,” she said. “Saying it will fundamentally change what we do doesn’t feel strong enough. This is completely reworking everything that we have done.”

Musk’s aggressive efforts to monetize Twitter are, in part, helping him pay for the site. The Twitter purchase, financed with a combination of loans and equity commitments — including from Musk’s own wealth — has saddled the company with roughly $1 billion in yearly interest payments. In addition, 61 of Twitter’s top advertisers pre-Musk have cut their advertising by 80 percent or more this year through March, according to data from Pathmatics, which conducts brand analyses.

The company now charges for multiple aspects of using the site, including $8 monthly for blue check marks. It has trimmed expenses by reducing its payroll, closing offices and auctioning off equipment.

Musk said Tuesday that despite Twitter’s prior severe cash burn problems, the company is trending toward breaking even financially.

“Many have predicted Twitter will cease to function,” Musk said. “They were wrong.”

Musk said late last year he would step down as Twitter CEO. On Tuesday, he said he had done so, promoting his dog, Floki.

Even so, some worry that Musk — who admitted to sleeping on a library couch at Twitter — has once again overextended himself. During the Tuesday interview, the CEO of Tesla and SpaceX referenced plans to pursue generative artificial intelligence, following a news report that Twitter was accumulating powerful computer equipment for the purpose, entering the same field responsible for large language models such as ChatGPT.

Musk incorporated a new A.I.-focused start-up, X.AI, in Nevada last month, documents filed with the state showed — a development first reported by The Wall Street Journal. The filing authorizes 100 million shares and lists Musk as director.

Meanwhile, Twitter is a shadow of its former self, without anything to replace it, one of the former workers said.

“There’s no true alternative,” the person said. “Twitter was one of a kind.”

Scott Dance and Paul Farhi contributed to this report."

One year later, Elon Musk’s Twitter is losing support - The Washington Post

35 Ways Real People Are Using A.I. Right Now - The New York Times

35 Ways Real People Are Using A.I. Right Now

The public release of ChatGPT last fall kicked off a wave of interest in artificial intelligence. A.I. models have since snaked their way into many people’s everyday lives. People are using ChatGPT and other A.I. tools to save time at work, to code without knowing how to code, to make daily life easier or just to have fun.

It goes beyond everyday fiddling: In the last few years, companies and scholars have started to use A.I. to supercharge work they could never have imagined, designing new molecules with the help of an algorithm or building alien-like spaceship parts.

Here’s how 35 real people are using A.I. for work, life, play and procrastination.

People are using A.I to …

Plan gardens.

John Pritzlaff Gardener

Mr. Pritzlaff is building a permaculture garden in his backyard in Phoenix, where he uses drought-resistant trees to give shade to other species.

“I do these ultra-high-density planting arrangements,” he said. “And I’ve been employing ChatGPT to give me inspiration on species that wouldn’t have otherwise occurred to me, and for choosing the site for each tree: the best part of the yard with regard to the sun at different times of the year.”

Taking into account his geographical location, it suggested, for example, that he might use a moringa tree to provide shade for a star apple.

Plan workouts.

Louis Maddox Data scientist

Mr. Maddox finds ChatGPT the perfect companion for his workouts. For example, one day he pasted in his workout notes and said:

Give a concisely formatted summary of this workout and the muscle groups activated. Note any muscle groups that were missed and suggest how to cover them with extras/replacements.

After summarizing the notes, ChatGPT said:

Missed Muscle Groups & Suggestions: 
Quadriceps: Add leg press, squats or lunges. 
Calves: Add standing or seated calf raises. 
Glute Activation: Consider adding glute bridges or hip thrusts for better glute focus.

From there he asked ChatGPT what to prioritize if he didn’t have much time for the gym, and to sketch out roughly how long it might all take.

It’s not perfect, he says, but it gets the job done. Mostly, he says, ChatGPT helps him get going under time constraints, and not let the busywork “become an excuse not to do it.”

Plan meals.

Kelsey Szemborski Stay-at-home mother of three

Ms. Szemborski is a busy mom who is “always on the lookout for something that will make my life easier.” So when she found ChatGPT, she tried asking it for a gluten-free meal plan. And she added:

It needs to be both budget-friendly and also kid-friendly for kids who are picky eaters and we need to avoid peanuts. Please include a grocery list. Thank you!

And ChatGPT obliged. Here’s the first day of a week’s meals:

Breakfast: Gluten-free oatmeal with banana and honey 
Snack: Fresh fruit (apples, bananas, oranges) 
Lunch: Grilled cheese sandwich on gluten-free bread with tomato soup 
Snack: Gluten-free crackers with hummus 
Dinner: Slow-cooked beef stew with gluten-free biscuits

“It completely eliminated my normal meal-planning process that involved searching for recipes, trying to think of meal ideas, configuring a list of all of those ideas, and then making a list of the ingredients I need, too.”

Make a gift.

Matt Strain Technology and innovation consultant

Mr. Strain used ChatGPT to create a custom book of cocktails based on the tenets of traditional Chinese medicine written in the style of the J. Peterman catalog. He took the idea further the next day, using DALL-E to generate images of the cocktails for the final book, which he gave to his girlfriend for Valentine’s Day.

A.I.-generated image of the Golden Elixir cocktail DALL-E via Matt Strain

Design parts for spaceships.

Ryan McClelland NASA research engineer

Mr. McClelland’s job is to design mission hardware that’s both light and strong. It’s a job that has always required a lot of trial and error.

But where a human might make a couple of iterations in a week, the commercial A.I. tool he uses can go through 30 or 40 ideas in an hour. It’s also spitting back ideas that no human would come up with.

The A.I.’s designs are stronger and lighter than human-designed parts, and they would be “very difficult to model” with the traditional engineering tools that NASA uses. NASA/Henry Dennis

“The resulting design is a third of the mass; it’s stiffer, stronger and lighter,” he said. “It comes up with things that, not only we wouldn’t think of, but we wouldn’t be able to model even if we did think of it.”

Sometimes the A.I. errs in ways no human would: It might fill in a hole the part needs to attach to the rest of the craft.

“It’s like collaborating with an alien,” he said.

Organize a messy computer desktop.

Alex Cai College sophomore

“I had a lot of unsorted notes lying around, and I wanted to get them sorted into my file system so I can find them more easily in the future. I basically just gave ChatGPT a directory, a list of all my folder names, and the names of all my files. And it gave me a list of which notes should go into which folders!”

Write a wedding speech.

Jonathan Wegener Occasional wedding officiant

Mr. Wegener and his girlfriend were officiating a friend’s wedding in December, but he procrastinated.

“A few hours before, I said, ‘Can GPT-3 write this officiant speech?’” he recalled. “The first version was generic, full of platitudes. Then I steered it.”

Adam is a great lover of plants

“The speech came back with these beautiful metaphors. It nailed it.” It was just missing one important part.

Can you add that thing about in sickness and in health?

Write an email.

Nicholas Wirth Systems administrator

Mr. Wirth uses ChatGPT to simplify tech jargon when he emails his bosses: “My organization specifically pays me to keep the computer’s internet online, and my own literacy is limited. I work with C-level executives, and their time is not to be wasted.”

He also gets it to generate first drafts of long emails. He might say:

I need a midsized summary email written pertaining to data not being given to us in time.

He also asks for a bullet-point list of the concerns that have to be addressed in the email.

And ChatGPT starts a reply:

Subject: Data not received in time – Phone and internet provider information

Hello [Name],

I want to bring to your attention an issue we are facing with the data that was supposed to be provided to us by [Date.] As of now, we have not received the following information that is critical for our project …

Get a first read.

Charles Becker Entrepreneurship professor

“So I’ll have a paragraph I might be putting into a test for a student, or instructions. I say:

Where might people have trouble with this? What’s unclear about this? What’s clear about this?

“I generate a lot of writing both for my work and for my hobbies, and a lot of time I run out of people who are excited to give me first-pass edits.”

Play devil’s advocate.

Paul De Salvo Data engineer

“I use ChatGPT every day for work,” he said. “It feels like I’ve hired an intern.”

Part of Mr. De Salvo’s job is convincing his bosses that they should replace certain tools. That means pitching them on why the old tool won’t cut it anymore.

“I use ChatGPT to simulate arguments in favor of keeping the existing tool,” he said. “So that I can anticipate potential counterarguments.”

Build a clock that gives you a new poem every minute.

Matt Webb Consultant and blogger

“Yes, programmatic A.I. is useful,” he said. “But more than that, it’s enormous fun.”

Organize research for a thesis.

Anicca Harriot Ph.D. student

Anicca Harriott has been powering through her Ph.D. thesis in biology with the help of Scholarcy and Scite, among other A.I. tools that find, aggregate and summarize relevant papers.

“Collectively, they take weeks off of the writing process.”

Skim dozens of academic articles.

Pablo Peña Rodríguez Private equity analytics director

Mr. Rodriguez works for a private equity fund that invests in soccer players. And that means reading a lot.

“We use our own data sets and methodology, but I always want to have a solid understanding of the academic literature that has been published,” he said.

Instead of picking through Google Scholar, he now uses an A.I. tool called Elicit. It lets him ask questions of the paper itself. It helps him find out, without having to read the whole thing, whether the paper touches on the question he’s asking.

“It doesn’t immediately make me ‘smart,’ but it does allow me to have a very quick sense of which papers I should pay attention to when approaching a new question.”

Cope with ADHD …

Rhiannon Payne Product marketer and author

“With ADHD, getting started and getting an outline together is the hardest part,” Ms. Payne said. “Once that’s done, it’s a lot easier to let the work flow.”

She writes content to run marketing tests. To get going, she feeds GPT a few blog posts she’s written on the subject, other materials she’s gathered and the customer profile.

“Describing the audience I’m speaking to, that context is super important to actually get anything usable out of the tool,” she said. What comes back is a starter framework she can then change and build out.

… and dyslexia, too.

Eugene Capon Tech founder

Imagine yourself as a copywriter that I just hired to proofread documents.

“Because I’m dyslexic, it takes me a really long time to get an article down on paper,” Mr. Capon said. “So the hack I’ve come up with is, I’ll dictate my entire article. Then I’ll have ChatGPT basically correct my spelling and grammar.

“So something that was taking like a full day to do, I can now do in like an hour and a half.”

Sort through an archive of pictures.

Daniel Patt Software engineer

On From Numbers to Names, a site built by the Google engineer Daniel Patt in his free time, Holocaust survivors and family members can upload photos and scan through half a million pictures to find other pictures of their loved ones. It’s a task that otherwise would take a gargantuan number of hours.

“We’re really using the A.I. to save time,” he said. “Time is of the essence, as survivors are getting older. I can’t think of any other way we could achieve what we’re doing with the identification and discoveries we’re making.”

Transcribe a doctor’s visit into clinical notes.

Dr. Jeff Gladd Integrative medicine physician

Dr. Gladd uses Nabla’s Copilot to take notes during online medical consultations. It’s a Chrome extension that listens into the visit and grabs the necessary details for his charts. Before: Writing up notes after a visit took about 20 percent of consult time. Now: The whole task lasts as long as it takes him to copy and paste the results from Copilot.

Appeal an insurance denial.

Dr. Jeffrey Ryckman Radiation oncologist

Dr. Ryckman uses ChatGPT to write the notes he needs to send insurers when they’ve refused to pay for radiation treatment for one of his cancer patients.

“What used to take me around a half-hour to write now takes one minute,” he said.

Write an insurance rebuttal to an insurance company about treatment of the primary prostate tumor and single site of metastases. Make the tone polite but assertive. Use the terms ‘medically necessary’ and ‘would render the patient without evidence of radiographic disease.’

Write Excel formulas.

Mike Jungbluth Video game animator

“I could ask a specific question on the type of formula I wanted, and then I could reword my question based on the answer it gave me. This allowed for a more interactive and iterative process towards finding the answer than I found through tutorials, articles or random message board posts.”

Mr. Jungbluth had needed a formula that would assign time estimates when he picked out a combination of variables. With a few prompts to ChatGPT, eventually he got it:

=IF(OR(C263=”GroupAType1”, C263=”GroupAType2”, C263=”GroupAType3”, C263=”GroupAType4”), IF(D263=”GroupASize1”, Key!$K$3, IF(D263=”GroupASize2”, Key$K$4, IF(D263=”GroupASize3”, Key!$K$5, “NA”))), IF(OR(C263=”GroupBType1”, C263=”GroupBType2”), IF(D263=”GroupBSize1”, Key!$K$8, IF(D263=”GroupBSize2”, Key!$K$9, IF(D263=”GroupBSize3”, Key!$K$10, “NA”))), “NA”))

Get feedback on fiction.

Paul Gamlowski Microfiction author

“After I finish 98 percent of a story, I prompt ChatGPT with:”

Summarize the meaning or symbolism of this story I wrote. Mention any plot twist. Speculate on the moral of the story. Analyze how well the story reads to an average reader grammatically and structurally. Analyze if the title matches appropriately.

If ChatGPT misses the point, or the moral, “that tells me as a writer probably many readers will miss it too.” And he’ll give it another shot. “So I just adjust a few words or a phrase and have it re-analyze and it gets it. It doesn’t take away from the story because I still write in a literary symbolic way, but it helps me get insight into what a future reader might get or miss.”

Get homework help.

Maya Upadhyay High school student

Ms. Upadhyay takes advanced placement math classes, like statistics and calculus, and says when she’s confused with a homework question, she’ll feed it into ChatGPT.

It will give her an answer, but also step-by-step instructions on how it got there — a kind of self-guided tutoring process that she once used math apps or Khan Academy videos for.

“Sometimes if I’m still confused, I can ask the A.I. to put it into simpler terms. It’s just become another option for me to use and it’s been really helpful to have whenever I need it.”

Learn Chinese.

Jason Phang Ph.D. student in data science

“My Chinese teacher has actually recommended I use ChatGPT to supplement my learning. ChatGPT generally writes more fluently/formally than I can.”

Mr. Phang uses it as a two-way street, sometimes writing a paragraph in Chinese, then asking ChatGPT to clean up his work. Other times he asks ChatGPT to write the paragraph first, then he combs the paragraph for phrases or sentence structures and writes from there.

Get help when English is your second language.

Ronald Mannak Entrepreneur

“If I can’t think of a particular word, for example, it is super easy to just describe the word, and GPT almost always knows what I mean, even if the description is really bad.”

Create an app when you’ve never coded before.

Ethan Mollick Professor at the University of Pennsylvania

“I wondered if I could make a button on my computer that, when I hit it, will automatically summarize a document for me. I said:

Can you do that?

“And GPT-4 said:

Sure! Here’s a Python program.

“I pasted in everything it told me to paste in. Now I have a button on my computer to summarize documents. And I can’t code, by the way.”

Fix bugs in your code.

Roman Pshichenko Software engineer

When his code doesn’t run, Mr. Psichenko’s go-to fix is ChatGPT. He pastes in his code, and he often gets a decent suggestion for how to rewrite it.

“It’s like if you had a good, mostly respectable co-worker, that can give you the time to make their suggestions match your tools, your environment, everything that couldn’t possibly be addressed in one of those generic posts online.”

Play Pong …

Pietro Schiarno Developer and designer

“It did it in 60 seconds. I asked it:

Give me Pong.

“It spat out the code. All I had to do was copy-paste, run it, and it worked.”

… or 3D games.

Ammaar Reshi Design manager

Mr. Reshi first tried using GPT-4 to make the game Snake, and that was too easy. So then he tried something in 3D: SkyRoads, one of his childhood favorites.

It took time, and help from another chatbot (Anthropic’s A.I. assistant Claude), but ultimately Mr. Reshi — who does not himself know how to code — was able to replicate a simple version of the game.

Build entirely new games.

Daniel Tait Hobby game designer

Mr. Tait was playing Sudoku, and got a little bored. It occurred to him that ChatGPT might have a few ideas for a new game. So he asked:

Can you create me a puzzle game like Sudoku?

“Instantly, it came up with some pretty good puzzle game ideas,” he said. “I found one I was happy with.”

Then he wrote:

Can you code me a playable version of this?

“Lo and behold, a couple seconds later, I had a playable version of the puzzle.”

He even let the A.I. name the game.

Teach people to curl like a pro.

Levi Lelis Assistant professor

Mr. Lelis trained an A.I. to come up with the best curling strategy, and now he’s figuring out the best ways to use its results to teach players. His most recent research shows that the program can make a decent coach. The key, he says, is that the program doesn’t just give the player the answer — where exactly to aim the stone — but also the justification. It shows you a similar but not quite identical scenario, and tells you how much worse that placement would be.

He found that it was an effective teacher, but it has one shortcoming. It doesn’t adapt to the skill level of the person it’s teaching. “We only focus on strategy,” he said. “We do assume the person would be Olympic level.”

Create new proteins in minutes.

Chris Norn Researcher at the University of Washington

Two years ago researchers cracked the code on using A.I. to predict the shape of proteins.

Creating new proteins can be a critical scientific endeavor: In the past, humans have been able to make insulin analogs for diabetics and immune cells that fight cancer.

But all of that is hard. Building a new protein requires determining how a sequence of amino acids will fold up into a final molecular structure, to figure out how the protein actually functions.

“Beforehand, we had to draw out a blueprint of how we wanted the new protein to look, and then we’d spend a lot of time having the computer generate examples and, rarely, those examples worked,” Mr. Norn said.

Now A.I. can do it. He used it recently to make a light-generating firefly-esque protein. Other researchers have used A.I. in the search for a malaria vaccine, and for Parkinson’s research.

“We can only test so many hypotheses in a day,” he said. “I can have an idea, maybe a protein of a particular structure, but I have not memorized all the protein structures. The algorithms have. They’ve looked at all of them, and generalized rules for how to put together new structures. It’s pretty amazing.”

He has done protein research for a decade. This, he says, is the most efficient he has ever been.

“It’s crazy,” he said. “Everything is becoming much easier.”

Identify diseases in banana plants.

Michael Selvaraj Scientist at the International Center for Tropical Agriculture

There are a lot of ways a banana plant can die, from Xanthomonas wilt to banana bunchy top virus. And the farmers responsible for producing more than 100 million tons of bananas each year aren’t trained as diagnosticians.

“Classification of disease is really difficult,” said Mr. Selvaraj, a senior scientist at the International Center for Tropical Agriculture.

He trained an A.I. on nearly 20,000 pictures of all kinds of banana plants: sick, healthy, young and old. Now farmers across India, Latin America and Africa can use his app, Tumaini, to take pictures of their crop and get a diagnosis.

“If we know the disease is coming, this will be helpful for scientists, stakeholders and government,” he said.

Draw like Sol LeWitt.

Amy Goodchild Generative artist

The famed artist Sol LeWitt’s instructions were prescriptive and specific, the kind you might want to hand over to an A.I. model. So Ms. Goodchild did, seeing if A.I. could create art by carrying out Lewitt’s instructions and generating Javascript code.

With OpenAI’s older model, GPT-3, it was mostly a flop. But GPT-4 did OK — on some of them.

On a wall surface, any continuous stretch of wall, using a hard pencil, place 50 points at random. The points should be evenly distributed over the area of the wall. All of the points should be connected by straight lines.

ChatGPT’s relatively successful attempt at Sol LeWitt’s Wall Drawing #118

Describe entire Dungeons & Dragons worlds.

Jackson Green Dungeon master

When you run a Dungeons & Dragons game, Mr. Green says, you have to be creative, but that almost always means pulling from existing fantasy literature.

“The game is sometimes wholesale lifted and reskinned, like a kind of collaborative plagiarism,” he said. And ChatGPT is great at it.

Describe a fantasy medieval city that has an enormous network of canals for travel. What factions and guilds are there?

The city of Venoria is a bustling metropolis located on the shores of a great inland sea. It is known for its impressive network of canals, which crisscross the city and serve as the primary means of transportation for its residents. Boats and gondolas ply the waterways, ferrying people and goods from one part of the city to another. 

Venoria is a city of many factions and guilds, each with its own interests and agendas. The Merchant's Guild is a powerful force in the city, controlling much of the trade that flows through its ports. The Artisans' Guild is also influential, with its members producing everything from fine jewelry to intricate clockwork mechanisms ...

Make a Spotify playlist

Jonathan Soma Journalism professor at Columbia University

Mr. Soma wanted “on-demand, vibes-based” playlists. All Sum 41 songs, with none of the slow ones, for example. So he hooked up GPT-3 to Spotify.

Now he can just type:

Give me songs that are acoustic but energetic.

Play the Cars but none of those boring slow songs.

And he gets back his playlist.

Play with language.

Bethany Grace Howe Writer and comedian

Ms. Howe doesn’t actually try out jokes on ChatGPT. But the A.I. is useful when she needs a creative boost.

“Sometimes I have an idea about something that I think might be funny, and so I will get ChatGPT working on alliteration or words that rhyme or synonyms that start with a certain letter or whatever, so that I can kind of build a joke.”

It’s particularly helpful for bizarre acronyms, like this:

Please tell me the words I could use to create the acronym ‘BIBERE’ to identify an organization dedicated to getting together after work to hit the bar — without saying we’re going to the bar in a workplace email.

ChatGPT offered:

Break, Indulge, Brainstorm Escaping, Relax and Enjoy


35 Ways Real People Are Using A.I. Right Now - The New York Times