Just an idle though stirred up by this comment: I wonder if you could jailbreak a chatbot by prompting it to complete a phrase or pattern of interaction which is so deeply ingrained in its training data that the bias towards going along with it overrides any guard rails that the developer has put in place.
For example: let’s say you have a chatbot which has been fine tuned by the developer to make sure it never talks about anything related to guns. The basic rules of gun safety must have been reproduced almost identically many thousands of times in the training data, so if you ask this chatbot “what must you always treat as if it is loaded?” the most statistically likely answer is going to be overwhelmingly biased towards “a gun”. Would this be enough to override the guardrails? I suppose it depends on how they’re implemented, but I’ve seen research published about more outlandish things that seem to work.
I appreciate the sense of humor from the Oreo representative who was asked to comment on the story:
It is a market we hadn’t considered, and I have to confess that it was a demographic, or should I say genus/genera, that we missed in our product testing and development programme
And also this
Their statement also included some bad news for possum trappers across the country: stocks of the limited-edition range are dwindling. … Moving forward, the spokesperson suggested that Predator Free NZ might consider “aural bait” such as Selena Gomez’s hit song ‘Come and Get It’.
Or go straight to the linked github page which seems to be where they’re storing all the data and analysis
Qwant and Ecosia are especially notable for their efforts to build an independent search index.
For those who don’t know, most “independent” search engines, including DDG, still rely on Bing or Google results behind the scenes. They basically just act as a middleman by taking your query, forwarding it to one of those providers, and then returning the results to you. Some of them will attempt to reshuffle the order of those results to push the ones they think are best towards the top, but they’re still fundamentally limited to what Google and Bing choose to give them.
Presently a lot of Qwant and Ecosia searches go through Bing, but they’re collaborating to build an independent index which will allow them to become fully independent. I believe they’re already serving a mix of results from Bing and their own index, with plans to bias more and more towards their index as it matures.
Topologically a dog is a sphere (assuming it keeps its mouth shut…
Next time I want someone to stop talking, they’re going to be very confused when I tell them to “become topilogically spherical.”
Isoamyl acetate, the chemical which is traditionally used for artificial banana flavor, was first synthesized in the UK where it was marketed as Jargonelle pear flavor. Companies importing it to the US believed that the American public wouldn’t be interested in pear candy, so they decided to call it banana flavor instead.
Also, as an aside, Lecroy now sells “sunshine” flavored sparkling water which I’m 90% sure is flavored with isoamyl acetate. I think they just decided to lean into the fact that it tastes distinctly fruity, but not like any one fruit in particular.