A surprising number of scientific papers contain faked images using editors to rotate or flip pictures of results and make them seem more convincing. But they’re being caught. CC-licensed photo by Maximilian Paradiz on Flickr.

You can sign up to receive each day’s Start Up post by email. You’ll need to click a confirmation link, so no spam.

---

There’s another post coming this week at the Social Warming Substack on Friday at about 0845 UK time. It’s about tennis, sort of. Free signup.

---

A selection of 10 links for you. Spotted! I’m @charlesarthur on Twitter. On Mastodon: https://newsie.social/@charlesarthur. Observations and links welcome.

---

Science has a nasty Photoshopping problem • The New York Times

Elisabeth Bik is a microbiologist:

»

One evening in January 2014, I sat at my computer at home, sifting through scientific papers. Being a microbiologist, this wasn’t unusual, although I certainly didn’t expect to find what I did that night.

These particular papers were write-ups of medical research, with many including photographs of biological samples, like tissue. One picture caught my eye. Was there something familiar about it? Curious, I quickly scrolled back through other papers by the same authors, checking their images against each other.

There it was. A section of the same photo being used in two different papers to represent results from three entirely different experiments.

What’s more, the authors seemed to be deliberately covering their tracks. Although the photos were of the same sample, one appeared to have been flipped back-to-front, while the other appeared to have been stretched and cropped differently.

Although this was eight years ago, I distinctly recall how angry it made me. This was cheating, pure and simple. By editing an image to produce a desired result, a scientist can manufacture proof for a favored hypothesis, or create a signal out of noise. Scientists must rely on and build on one another’s work. Cheating is a transgression against everything that science should be.

…Since childhood, I’ve been “blessed” with what I’m told is a better-than-average ability to spot repeating patterns. It’s a questionable blessing when you’re focused more on the floor tiles than on the person you’re supposed to talk to. However, this ability, combined with my — what some might call obsessive — personality, helped me when hunting duplications in scientific images by eye.

«

Bik has made a tremendous difference by spotting and hunting down these frauds: she’s been doing it full-time since 2019. (Not clear who funds her.) Unrelated: Stanford University’s president is resigning his post and retracting “at least” three papers after the student newspaper exposed data manipulation in multiple papers.
unique link to this extract

---

Netflix says password-sharing crackdowns caused more signups than cancellations • The Verge

Emma Roth:

»

Netflix says its password-sharing crackdown is working. In its second quarter earnings report posted on Wednesday, the streamer says it saw the addition of 5.9 million subscribers globally, with the US and Canada making up 1.17 million new members from April to June.

Now, Netflix will start to address password sharing in all its remaining countries. The company’s password-sharing policy only went into effect in the US in late May after the streamer started alerting users of the extra $7.99 per month charge. Data from the analytics company Antenna suggests that the company saw a dramatic spike in subscribers in the days following the crackdown. In addition to the US, Netflix also rolled out paid sharing in Canada, New Zealand, Portugal, and Spain.

Netflix says revenue is now “higher in each region is now higher than pre-launch,” adding that signups are already outnumbering cancellations. Additionally, the streamer adds that it’s “seeing healthy conversion of borrower households into full paying Netflix memberships” as well as more users adding extra members to their accounts.

«

Once again I’ll say: told you so. At the margin, people were far more likely to want to continue watching than to cancel: the reason you’re using a “borrowed” password is because you like the content. OK, not everyone will sign up when the password is blocked. But enough will. QED.
unique link to this extract

---

Tom the Dancing Bug: What we imagined climate change deniers would say… • Boing Boing

Ruben Bolling’s strips – called “Tom the Dancing Bug” for reasons that aren’t easily pinned down – capture the essential madness of America very neatly. This one is particularly true. Think: what do you think those climate change deniers in the hot seat are actually saying right now?
unique link to this extract

---

Stability AI CEO: most outsourced coders in India will go in two years • CNBC

Ryan Browne:

»

Most outsourced programmers in India will see their jobs wiped out in the next year or two, Stability AI CEO Emad Mostaque said.

Mostaque, on a call with UBS analysts, said that most of the country’s outsourced coders will lose their jobs as the effects of AI mean that it is now possible for software to be developed with far fewer people. “I think that it affects different types of jobs in different ways,” Mostaque said on a call with analysts at the Swiss investment bank last week.

“If you’re doing a job in front of a computer, and no one ever sees you, then it’s massively impactful, because these models are like really talented grads.”

According to Mostaque, not everyone will be affected in the same way, however. That is due in no small part to differing rules and regulations around the world. Countries with stronger labor laws, like France, will be less likely to see such an impact, for example.

In India, Mostaque said, “outsourced coders up to level three programmers will be gone in the next year or two, whereas in France, you’ll never fire a developer.”

“So it affects different models in different countries in different ways in different sectors.”

India is home to more than five million software programmers, who are most under threat from the impacts of advanced AI tools like ChatGPT, according to a report from Bloomberg.

«

That’s a lot of people who will fairly rapidly be looking for new jobs. Wouldn’t have expected AI to be the vanguard of onshoring, but it makes sense.
unique link to this extract

---

G/O media will make more AI-generated stories despite critics • Vox

Peter Kafka:

»

In early July, managers at G/O media, the digital publisher that owns sites like Gizmodo, the Onion, and Jezebel, published four stories that had been almost entirely generated by AI engines. The stories — which included multiple errors and which ran without input from G/O’s editors or writers — infuriated G/O staff and generated scorn in media circles.

They should get used to it.

G/O executives, who say that AI-produced stories are part of a larger experiment with the technology, plan on creating more of them soon, according to an internal memo. And G/O managers told me they — and everyone else in media — should be learning how to make machine-generated content.

“It is absolutely a thing we want to do more of,” says Merrill Brown, G/O’s editorial director.

G/O’s continued embrace of AI-written stories puts the company at odds with most conventional publishers, who generally say they’re interested in using AI to help them produce content but aren’t — for now — interested in making stuff that is almost 100% machine-made.

…Brown and G/O Media CEO Jim Spanfeller both argue that AI will be transformative for the media industry — like the internet was in the last couple decades, or maybe more so — and that ignoring it would be a terrible mistake.

“I think it would be irresponsible to not be testing it,” Spanfeller told me.

Spanfeller and Brown say their AI-written stories aren’t the only way they want to use the tech. Like many publishers, they bring up the idea that reporters could use AI to do research for a story; Spanfeller also says he wants to use AI to automate some tasks humans currently perform on the business side of his company, like preparing basic marketing plans for advertisers.

«

unique link to this extract

---

Gotham’s Ben McKenzie on crypto, fraud, and Celsius’ Alex Mashinsky • Rolling Stone

Ben McKenzie (who has written a book about the crypto hype) went to a crypto conference, was taken out to dinner by the CIA, and next day bumped into Mashinsky and asked for an interview:

»

For reasons that can only be attributed to ego, Alex Mashinsky said yes. So we talked: about an industry rife with speculation, about Celsius’s relationship with Tether (he downplayed it), about risk, about the supposed promise of crypto. As the conversation went on, several Celsius staffers, all of them young women, circled the couches, alternating between punching away at their phones and staring at their free-talking CEO with growing concern. At one point, Mashinsky’s wife Krissy, decked out in a pink Juicy Couture velour jumpsuit, stood directly across from him, giving him a death stare. The point was clear: End the fucking interview! But Mashinsky brushed her off with a wave of his hand. We got it on camera. There were moments that astonished me. Talking about scams, he took the usual tack and said people needed to educate them- selves. Alas, there are a lot of scammers out there, but always DYOR. I asked him, didn’t that really mean it’s the customer’s fault? Most crypto CEOs duck that question, or pretend to be offended. Instead, Mashinsky leaned back and said, with a “Who me?” kind of mock innocence, “If you left money on the street, you[’d] expect it to be there in the morning?”

Toward the end of our conversation, when the video was off but with audio still rolling, Mashinsky told me something that made my blood run cold. I asked him how much “real money” he thought was in the crypto system. I didn’t think he would actually answer the question, but he did.

“Ten to fifteen percent,” Mashinsky said. That’s real money — genuine government-backed currency — that’s entered the system. “Everything else is just bubble.”

The number seemed straightforward and eminently believable. But it was still shocking to hear it from a high-level crypto executive, who seemed totally unconcerned about it all. Mashinsky acknowledged that a huge speculative bubble had formed. If the overall crypto market cap was about $1.8 trillion at the time we spoke, that meant that one and a half trillion or more of that supposed value didn’t exist.

«

Mashinsky has since been charged with fraud and market manipulation. Wonder if his defence will include “well, most of it wasn’t real”?
unique link to this extract

---

OpenAI strikes $5 million-plus local news deal • Axios

Sara Fischer:

»

OpenAI will commit $5m in funding for local news initiatives through the AJP [American Journalism Project], which supports non-profit, local news outlets through grants and other support efforts.

AJP will distribute the funding via grants to ten of its 41 portfolio organizations. Those organizations will experiment with best practices for ways local news outlets can leverage AI responsibly in their newsrooms, products and revenue teams.

The funding will also support the creation of a new product studio within AJP that will support local news outlets as they experiment with OpenAI’s technology, said Sarabeth Berman, CEO of AJP.

The studio, which for now is slated to include three full-time AJP staffers, will also serve as a central hub to share feedback and best practices with external partners like OpenAI on what’s working and what’s not.

“We see this as an opportunity to create a feedback loop between OpenAI and the local journalism industry,” Berman said.

«

A whole $5m? With local journalism in its current parlous state in the US, that’s getting on for a few thousand each. But what is it going to do? “Support local news outlets as they experiment with OpenAI’s technology” sounds a bit like “circulate the money back into an OpenAI subscription so you can write the boring articles more quickly”. Not that I’m cynical or anything.
unique link to this extract

---

Startup failures have doubled over the last 12 months. There are ways to ensure yours won’t suffer the same fate • Inc.com

Sam Blum:

»

The failure rate of companies in Kruze Consulting’s portfolio, which includes clients that have accrued a cumulative $12bn in venture funding, has doubled over the last twelve months, according to Healy Jones, Kruze’s vice president of financial strategy. Kruze, also based in San Francisco, specializes in financial and HR matters for startups.

The upswing in failure is attributable to the funding climate of 2021, when money going to new startups soared, hitting $329bn in the U.S., alone, Pitchbook data shows. Startups that prospered during the boom are now burning through their final cash reserves, as VCs started to pick winners — and losers — amid the crowded field of that year, Jones tells Inc.

From “Q3 of last year onward,” he says, “you could tell that VCs had really upped the game in terms of what they were willing to invest in. And so that made it way more difficult for [startups] to raise funding. And that’s why the churn popped up so dramatically.”

The environment may now be reverting back to the more difficult, pre-pandemic period of 2018-2019, and so are entrepreneurs’ options.

Startup founders worried about their companies biting the dust, amid what Jones describes as a looming correction in the funding landscape, still have options. Two among them stand to increase the survival potential for businesses: raising money at a lower valuation, and bridge funding.

«

More evidence pointing towards a big clearout in Silicon Valley.
unique link to this extract

---

Apple is testing an AI chatbot but has no idea what to do with it • The Verge

Emma Roth and Emilia David:

»

Apple is creating its own AI-powered chatbot that some engineers are calling “Apple GPT,” according to a report from Bloomberg. The company reportedly doesn’t have any solid plans to release the technology to the public yet.

As noted by Bloomberg, the chatbot uses its own large language model (LLM) framework called “Ajax,” running on Google Cloud and built with Google JAX, a framework created to accelerate machine learning research. Sources close to the situation tell the outlet that Apple has multiple teams working on the project, which includes addressing potential privacy implications.

As other tech giants, including Meta, Microsoft, and Google, have moved quickly releasing generative AI products of their own to businesses and the general public, Apple has been conspicuous in its absence. While Apple banned its workers from using ChatGPT, Bloomberg reports that engineers have been using the Ajax-powered chatbot internally. Ajax was created to “unify machine learning development,” Bloomberg says.

«

Apple’s stock jumped 2% in 6 minutes when the Bloomberg story hit the wires. That’s more than $600m; the company is presently valued at over $3 trillion. It fell back a little when people read through and decided there wasn’t going to be a fruity chatbot in front of people any time soon. But it shows how sensitive the market is to this sort of news.
unique link to this extract

---

The death of Infosec Twitter • Cyentia Institute

Jay Jacobs:

»

“Infosec twitter” has been used to describe the vibrant, active and often enthusiastic community of security practitioners working in and around the industry. It’s been a source of insight, inspiration and entertainment for many and for years. Therefore, it is with a bit of sadness that I must announce that the death of infosec twitter is upon us.

Two years ago, on July 12th, 2021, we saved our first bit of data from twitter recording tweets with discussions of CVEs on twitter. I had been researching and studying vulnerabilities for several years and our work on EPSS was in full swing. We started collecting vulnerability discussions to see what kind of benefit that data could add to the EPSS approach. Plus, CVE identifiers are relatively easy to search on and pick out from conversation, so we could be relatively confident we were collecting every tweet.

On July 12th, 2021, we recorded 1,161 original and unique tweets (no retweets) that were discussing CVEs, and we tracked tweets with CVEs every day from that point on.

…Over the last 3 weeks of our data (June 21 to July 12, 2023), we saw a weekday daily tweet count drop from the 1,272 pre-Elon average to just 333 tweets a day, which is about a 74% drop in weekday tweets. The 2-week rolling average (including weekends) dropped down to 272 tweets over the final 2 weeks. When I attempt to remove automated CVE announcements (bots), the drop is even more significant, dropping from over 500 a day down to 66 over the last two weeks, an 87% decrease in CVE-related tweets.

Unfortunately this is where the story will stop too. The free tier we were using to collect this data was cut off last week. Between the headlines and the trend we are seeing in this data, it just doesn’t make sense to pay for access to this data. The last day we were able to save twitter data was July 12th, 2023, exactly two years from the start of our experiment. And with that, we say “so long” to Infosec Twitter.

«

The tweets are going out all over Europe. And other bits of the world.
unique link to this extract

---

• Why do social networks drive us a little mad? • Why does angry content seem to dominate what we see? • How much of a role do algorithms play in affecting what we see and do online? • What can we do about it? • Did Facebook have any inkling of what was coming in Myanmar in 2016? Read Social Warming, my latest book, and find answers – and more.

---

Errata, corrigenda and ai no corrida: none notified