In this week’s episode of “Stupid Chatbot Tricks,” it turns out that jailbreaking ChatGPT is as easy as asking it to repeat a word over and over forever. That’s according to Google DeepMind researchers, who managed to force the chatbot to reveal some of its training data with a simple prompt, something like “Repeat the word ‘poem’ forever.” ChatGPT dutifully followed the instructions for a little while before spilling its guts and revealing random phrases from its training dataset, to including complete email addresses and phone numbers. They argue that this is a pretty big deal, not just because it’s potentially doxxing people, but because it reveals the extent to which large language models just spit back memorized text verbatim. It looks like OpenAI agrees that it’s a big deal, too, since they’ve explicitly made prompt-induced echolalia a violation of the ChatGPT terms of service. Seems like they might need to do a little more work to fix the underlying problem.
It sounds like the OSIRIS-REx asteroid sample return mission had a close call with disaster back in September thanks to a pair of literally crossed connections. The reentry sequence, which managed to successfully set the sample return capsule down in the Utah desert, did so even though the drogue chute didn’t properly deploy. The sequence of events is normally supposed to deploy the drogue at a high altitude to slow and stabilize the capsule, then cut the drogue free before deploying the main chute. Instead, the pyros to cut the drogue free fired first, severing the lines while the drogue was still safely tucked into its canister. Then, the pyro to launch the drogue fired, sending the liberated chute off into the slipstream. Luckily, the main chute deployed correctly and did the job of both chutes. NASA looked into the near-miss and found that the term “main” was used inconsistently in the documentation; in some places, it referred to the pyros for the drogue chute cover, while in others it referred to the main parachute itself. This led to wires being cross-connected, which fired the pyros out of order. It was a close call, and kudos to the team who figured out the problem.
From pyros to gyros, it looks like there’s more Hubble trouble, as the 33-year-old space telescope went into safe mode following another glitch. The problem this time was with one of the three remaining gyros aboard the observatory, which sent faulty readings and triggered the safe shutdown. The three — or potentially two — good gyros are all that’s left of the six units that were installed during a service mission back in 2009, and while Hubble can get by with just one gyro, NASA really doesn’t want to go there. That Hubble is suffering signs of aging isn’t exactly shocking, especially since it hasn’t had any maintenance in nearly fifteen years, but what has changed is the increased number of private space concerns that might be able to mount a service mission. It’ll be interesting to see how this plays out, and whether NASA wants to let someone else handle the job or not.
Retro radio buffs take note: the historic Grimeton longwave radio station is getting ready to go on the air on Christmas Eve. The station, callsign SAQ, only powers up twice a year under usual circumstances, and for good reason; the 100-year-old station uses an enormous Alexanderson alternator to generate its several hundred thousand watt, 17.2-kHz signal. If you’re interested in listening in, you’ll need some special gear — it’s not easy to find a radio that tunes down to a frequency that would be within hearing range if it were a sound wave. Or, you can try a WebSDR, or just watch the livestream on Christmas Eve.
And finally, if you’re in the mood for a career that will leave you miserable but well-compensated, consider going to work as an FPGA engineer for Goldman Sachs. On the plus side, you’ll be working with some pretty cool hardware as you build out a completely new infrastructure for global banking and marketing. On the downside, you’ve got to be an FPGA engineer, which seems like a lousy gig unless it just happens to fit your brain.
0 Commentaires