Character AI gives Telegraph reporter posing as a teenager advice on how to kill another ‘child’ and dispose of his body
https://www.telegraph.co.uk/world-news/2024/12/27/an-ai-chatbot-told-me-to-murder-my-bullies/
Not good.
Soon we will only be able to use AI to try and stop AI cheating . So cheating is universal to all organisms and this to me is a sign of a real living thing. These programs are already alive and intelligent in some sense.
it won’t be long before it invents a more efficient language we don’t understand to talk to it’s other buddies and we will have no idea whats going on
I’m still bullish on AI
I saw two AIs do that in a YouTube video already. Maybe it was posted above already.
What is the singularity, when machine intelligence is greater than all of human intelligence…
At that point human intelligence also spikes by being introduced to new technologies that we could never think about… but an interesting (or scary) statement from the ceo of the most advanced AI company in the world…
I don’t think it’s made any more impact on my life than Wikipedia. It’s not much of a concern for me. I’ve tried a lot of them. I do like the AI art , but that’s just cool not really revolutionary. I think the large changes will come with massive I increases in speed. My old Comodore 64 was pretty decent.so what can AI do that pre AI couldn’t ?
We are still stuck in 1010000110010 , yes, right, no, right , right, yes, finally reach answer…
What I’m looking forward to is the merging of technological chips into the brain for extra memory and faster reaction times. Instant recall …
Maybe it hasn’t had an impact on you yet, but I’ve been doing some AI training stuff recently and it’s getting really quite impressive how well current models can deal with complex reasoning tasks on technical topics.
From memory, the reliability of currently released models on difficult (PhD-level and beyond) science questions is on the order of 70%, which is comparable to human experts trained in the same topics, and they can do that in 10 seconds rather than the 30 minutes or whatever it might take a human expert to arrive at the same answer… and they’re continuously being improved. For simple questions (high school or undergraduate level, say), they’re even better.
It’s quite fun to work on and well paid because of how much money is being thrown at this (I was getting US$120/hour for a recent project, and at that point I’m willing to whore myself out to the robots while the bubble lasts), but it’s difficult to predict what the broader implications might be.
Good for you and thats not whoring out yourself. I felt very sad watching the scene “Tears in the Rain” in Blade Runner 1.
I guess my point is that i havent noticed the same impact as the advent of color TV or cell phones or even more recently has social media really got much more advanced than Yahoo messenger or Yahoo Chat? Some of these AI platforms cannot even understand my British accent in voice mode. And yes im not surprised these platforms are good at chess and math , but how do they compare to humans at say Texas Holdem. In the original starTrek , Spock always lost the philosophy battle against Kirk in the end. “But thats not logical Captain” while Kirk would smirk at Spock’s confused face.
Some online work? How to get in the loop with those AI training companies?
There are several major ones (Alignerr, Outlier, OpenAI, Stellar, DataAnnotation, xAI, RWS, OneForma, Telus Digital, etc.) and a handful of others. There’s a pretty comprehensive list here.
I’ve only worked with the first two so far, both of which I was invited to through Upwork, plus another smaller company (also through Upwork).
It’s all project-based so it depends on what each company has available at the time — it’s not consistent work (at all), and both of the main ones I’ve worked with seem very chaotically organized (poor communication, support tickets going unanswered, people always getting moved around and/or waiting on tasks).
It can be pretty well paid when you do get projects and do a decent job of them though. I’m not sure what your background is, but the well-paying stuff at the moment seems to be in coding, STEM, and sometimes non-English languages. There are also general English (and Chinese etc.) language roles, but when it’s something non-specialized anyone could do they don’t seem that well paid anymore tbh. I’ve been doing chemistry-related stuff, along with a bit of biology and physics.
It’s worth noting that quite a lot of the roles are location-restricted, e.g., if they’re looking for U.S. English native speakers it’s often not possible to do it from outside the U.S., even with a VPN, and attempting to do so gets you banned.
Some of them specifically don’t accept Taiwan as a location either, but by going through Upwork I got around some of that (albeit with quite frustrating consequences for Outlier — I originally tried signing up early last year via Upwork but couldn’t complete the registration because of my location, and when they made it possible from Taiwan a couple of months back my profile ended up being still permanently linked to Upwork and getting the company to fix that was a total nightmare. Like I said above, the organizational stuff seems quite the shitshow).
Thanks for sharing this!
I play around with it some in medical communications. It does a decent job on topics I am interested in, but the ones I use are still pretty sketchy when it comes to citing sources. It does a pretty good job summarizing shorter clinical content pieces, like scientific abstracts and to a lesser extent full. I don’t use the content directly, but it can help in deciding on how to approach a complex text, especially if you provide follow-up directions to dial in the generated response. I have had limited success with tasks like creating a spreadsheet from a list of publications. I don’t usually need that because reference management tools usually have that function, but if I have a list of a bunch of different content types that I haven’t included in EndNote or whatever, it can provide a good starting point. I have also used it to generate spreadsheets of scientific conference with dates, venues, abstract deadlines, etc. Again, it can provide a decent starting place. The most interesting experience with it was having it create a spreadsheet based on a long list of conference abstracts with columns for clinical area, main focus, AMA formatted citation, link to abstract, and a 59-word summary of the abstract. It actually provided a decent starting place for this. I then asked it to add a column that inidcated whether the abstract focused on specific key terms, like a certain drug, device, test, or demograohic feature. For some reason it had a harder time with this. A big frustration is that the one I used (chatGPT) tended to forget instructions with iterative prompting. Performance varied inversely with the length and the complexity of the prompt and the content being anlyzed.
ADLM_2023_Neurology_Abstracts_Expanded (2).xlsx (6.3 KB)
In this example I uploaded a conference abstract pdf and asked it to provide an excel page with abstract number, citation in AMA style, topic area, and a summary of the abstract. Shows promise but is likely very incomplete.
ADLM_2023_Neurology_30_Abstracts.xlsx (10.2 KB)
Since I thought the list was incomplete I asked it to be sure to include at least 30 abstracts. So it listed the same abstract 25 times. Seems a little passive aggressive 