| Hi, it's Drake in New York. The AI hallucination problem is more complicated than it seems. But first... Three things you need to know today: • OpenAI is allowing employees to sell shares
• Amkor plans to build a semiconductor facility in Arizona
• It's been one year since ChatGPT launched For office workers, the list of performance-enhancing drugs is pretty short. There's caffeine, of course, and Adderall for the hardcore. For members of the burgeoning microdosing community, there are psychedelics. Microdosing is the practice of taking small, almost homeopathic doses of LSD or psilocybin or other hallucinogenic compounds, below the threshold that would trigger a full-blown trip. Its adherents — well-represented among the life hackers and psychonauts of Silicon Valley — swear that it improves their mood and alertness and creativity while allowing them to spend a day productively coding or composing music or dreaming up outside-the-box ad taglines without wigging out their co-workers. There isn't a lot of medical research into the activity, which isn't surprising since the compounds remain illegal in most places. But all of that is changing. As laws against the drugs soften and as research suggests the compounds at full dose can have dramatic benefits for people with PTSD and other disorders and as an industry grows up to take advantage of that, microdosing is getting a more quantitative look, too. Now, a team of European researchers has published a preprint (not yet peer-reviewed, in other words) of a paper describing an experiment they carried out at a microdosing event hosted by the Dutch Psychedelic Society. The researchers gave attendees a series of cognitive tests before and after they microdosed on psychedelic truffles. The study found that, while so-called fluid intelligence — a measure of abstract reasoning — didn't improve, two measures of creativity did show positive effects. One of those was convergent thinking, which the paper's authors define as "identification of a single solution to a well-defined problem." For example, asking someone what word could be combined with "spot," "flower" and "set." ("Sun," get it?) The other measure that improved was divergent thinking, or "the collection of many possible solutions to a loosely defined problem." Like, asking how many different things you could use a brick for. The authors aren't sure why the drugs would improve both of those measures at once — after all, the two are somewhat at odds. And it's important to repeat that the study hasn't been peer-reviewed. Also, there wasn't a control group, a decision the authors defend on "ethical and practical reasons" — presumably it would have been a bit mean to show up at the Dutch Psychedelic Society's microdosing event and announce that half the attendees had to take placebos. Still, the results are intriguing and not just for people considering incorporating microdosing into their work regimen. In a way, they have a bearing on some of the discussions around artificial intelligence. AI "hallucinations" — when an AI simply makes something up about the world — are mostly seen as a problem to be solved. They're certainly a problem if you want an AI to give you tax advice or diagnose an illness. But maybe not if you want it to be creative. As Sam Altman himself pointed out in a recent interview, imagination — at the margins — is hard to distinguish from hallucination. So, for certain purposes, we might want AI that hallucinates. Maybe not full-fledged hallucination, but something in low-earth orbit around reality. To be creative, AI probably doesn't need to trip, but maybe it needs to microdose. —Drake Bennett In the year since ChatGPT debuted, companies are on the hunt to hire so-called prompt engineers. It's one of the unexpected new jobs that has resulted from the AI boom. Startups had a rough year, marked by mass layoffs, down rounds and bankruptcies. Apple faces a revived probe into mobile browsers and gaming in the UK. Meta sued the FTC, claiming the agency's in-house court violates the US Constitution. Tech leaders pushed the TikTok CEO on moderation of antisemitic content. | 
No comments:
Post a Comment