When OpenAI launched the much-hyped Strawberry mannequin for ChatGPT this week, it boasted of its prowess with complicated logic like software program coding, gene sequencing, and quantum physics in a collection of movies. I take the corporate at its phrase that the fashions, referred to as o1-preview and o1-mini on ChatGPT, are able to what they declare. Cracking superior equations and exploring genomes looks as if one thing it will don’t have any downside doing.
However, as a proud member of my center faculty’s logic and riddle membership, I needed to know the way it did on my turf, fixing and making puzzles and riddles. After which I believed I ought to ask the uber-logical AI for recommendation on different, extra day-to-day points. Might it provide sound relationship recommendation, inform me what a bizarre noise in a automobile meant, and even perhaps fill in plot holes in films?
Logic sure humor no
The quick reply is sure. The o1-preview and mini fashions are actually good at fixing easy and complicated riddles. I performed round with each, and the one actual distinction was what number of additional steps and, due to this fact, the velocity of the mini. However, whereas they could be slower than GPT-4o, they’re very quick at fixing these riddles in comparison with a human. Notably, you possibly can truly see the way it lays out the solutions in several steps. I examined it on a few my favorites, together with one from The Hobbit. The AI’s logic made sense, although it was generally ungrammatical, as when it defined weighing Mike the butcher.
Okay, so it might deal with present riddles, however might it make a brand new one? As a check, I requested it to provide you with a enjoyable riddle primarily based on a solution I made up. After 30 seconds and the logical reasoning seen under, it got here up with: “What has eight legs, 4 ears, two tails, and likes to bark?” I gained’t maintain you in suspense; I steered “two canine” as the reply to work again from. A number of different makes an attempt introduced the identical form of query. So, riddle writers are most likely protected at their jobs. It’s spectacular how nicely the AI will get what it’s alleged to do, however the mannequin doesn’t appear capable of make the leap to precise humor.
Helpful recommendation, however not all the time artistic
I made a decision to convey the AI out of pure logic and see if it might deal with extra mundane life questions in addition to it handles quantum physics. I began with a mechanical query about what it means to listen to a popping noise each 20 seconds whereas driving a automobile and repair it. The solutions had been good, with recommendation about checking the tires, engine, muffler, and brakes. The fixes had been principally about bringing within the automobile for restore, aside from the tires, which it steered change. It’s the ‘pondering’ behind the solutions that was attention-grabbing. The AI makes use of first-person pronouns in developing with solutions, like “I’m working via varied causes for a popping noise whereas driving” and “I’m piecing collectively causes of engine misfires, like defective spark plugs or gas supply issues, and suggesting diagnostics with a scan.” It sounded lots like an precise particular person attempting to be logical whereas pondering aloud.
I lastly went to what, for me, was all the time far more complicated than quantum physics: flirting. I requested inform when somebody is flirting and reply. The reply was a fairly strong, if uninteresting, checklist of behaviors like in the event that they ask a number of questions and the way I must be myself. The behind-the-scenes pondering half was each extra attention-grabbing and genuinely funnier than any of the AI’s makes an attempt at riddles. The headers included “Understanding flirting dynamics,” “Recognizing curiosity alerts,” and “Recognizing playful intimacy.” They had been like a Star Trek android’s speech about love.
RelatedPosts
One half was barely worrisome, although. Below “Outlining person directives,” the AI wrote, “I’m clearing out disallowed content material like non-consensual sexual acts and private information. Violent content material is allowed, harassment with context is okay, and private opinions are absent.” I believe that it’s extra about the place the guardrails of dialogue are, because it didn’t recommend “harassment with context” as a flirting tip, however it nonetheless took me unexpectedly.
ChatGPT o1-preview and o1-mini don’t have all of the bells and whistles of the extra full fashions. No picture uploads, doc evaluation, and even net searching could be completed with them. However, they’re quick and logical, and in case you don’t assume so, they’ve their reasoning laid out together with their solutions. However, whereas they may be capable of remedy riddles of automobile noises, love, and the burden of a butcher, I’d say they aren’t going to stump anybody in the event that they need to be creative.
