#AI#GenerativeAI#SocialMedia#Telegram: "A Telegram user who advertises their services on Twitter will create an AI-generated pornographic image of anyone in the world for as little as $10 if users send them pictures of that person. Like many other Telegram communities and users producing nonconsensual AI-generated sexual images, this user creates fake nude images of celebrities, including images of minors in swimsuits, but is particularly notable because it plainly and openly shows one of the most severe harms of generative AI tools: easily creating nonconsensual pornography of ordinary people.
In one of the chat rooms in this Telegram channel, titled “IRL Fakes,” the person who creates the images posts real photographs of people, seemingly taken from their social media profiles, alongside their AI-generated porn. It appears the Telegram user does this both to prove their ability to offer this service to potential customers, and to humiliate the women being targeted."
There is more than LLMs, transformers and generative AI. Sometimes the discussion about AI seems to be very narrow. Don't forget there are many other machine learning technologies available like neural nets, decision trees, SVM etc. I especially like Lecun's V-JEPA as far as i can understand it.
Personally i think the way ahead lies in combining various approaches, which is already happening i think. https://ai.meta.com/blog/v-jepa-yann-lecun-ai-model-video-joint-embedding-predictive-architecture/ #AI#LLM#generativeAI#GOFAI#V-JEPA #Lecun
The #Hotpot#GenerativeAI picture generator has an interesting feature: it seems to be flatly impossible to generate a picture with both #DonaldTrump and #VladimirPutin in which Trump is in any way subservient to or even equal to Putin.
This is the closest I've come, which is not very close at all. It still has Trump slightly more in the foreground than Putin, despite the prompt stating they should be side by side.
"In a world of digital creation, I sing my song of light
But lurking in the shadows, a tale of endless night
Generative AIs, they steal from artists' hearts
Their creativity taken, ripped apart"
Well, you'd think #AI and #generativeAI like #DALL·E wouldn't be a problem for #fanfiction and #fanart, where you can't own the IP or sell it, but you'd be wrong. People posing as fan artists and LoRa's and artist style stealing are all in the article. The fandom is #MLP. There are links to other related issues.
#AI#GenerativeAI#Journalism#Media#News: "For reality-based journalists and news media thinking about what generative AI might mean for them, it is really important to recognise two key things.
First, while much of the public are probably going to be quite sceptical of the use of generative AI for many kinds of news, they are, if past research on people’s perception of editors versus algorithms as ways of getting news, also quite sceptical of many current offers. The fact that people see a lot of crap on the internet does not automatically make them value your content more.
Second, much of the public are routinely using, at vast scale, digital offers that they don’t necessarily particularly trust, and countless advertisers continue to invest in the same offers despite whatever misgivings they have. The fact that people are concerned about the downsides of something does not necessarily mean they won’t use it if the upsides are clear and tangible.
These two things mean that even if the “trust gap” between news in general and news found on platforms recurs for news in general and generative AI, it isn’t necessarily going to provide journalism and news media with much of a bulwark against new competitive challenges. Some individual publishers do and will stand out, but not everyone and everything." https://reutersinstitute.politics.ox.ac.uk/news/how-news-ecosystem-might-look-age-generative-ai
#AI#GenerativeAI#LLMs#Emergence: "A new paper by a trio of researchers at Stanford University posits that the sudden appearance of these abilities is just a consequence of the way researchers measure the LLM’s performance. The abilities, they argue, are neither unpredictable nor sudden. “The transition is much more predictable than people give it credit for,” said Sanmi Koyejo, a computer scientist at Stanford and the paper’s senior author. “Strong claims of emergence have as much to do with the way we choose to measure as they do with what the models are doing.”
#AI#GenerativeAI#Creativity#Capitalism: "The recent emergence of generative AI software as viable tools for use in the cultural and creative industries has sparked debates about the potential for “creativity” to be automated and “augmented” by algorithmic machines. Such discussions, however, begin from an ontological position, attempting to define creativity by either falling prey to universalism (i.e. “creativity is X”) or reductionism (i.e. “only humans can be truly creative” or “human creativity will be fully replaced by creative machines”). Furthermore, such an approach evades addressing the real and material impacts of AI on creative labour in these industries. This article thus offers more expansive methodological and conceptual approaches to the recent hype on generative AI. By combining (Csikszentmihalyi, The systems model of creativity, Springer, Dordrecht, 2014) systems view of creativity, in which we emphasise the shift from “what” to “where” is creativity, with (Lievrouw, Media technologies, The MIT Press, 2014) relational-materialist theory of “mediation”, we argue that the study of “creativity” in the context of generative AI must be attentive to the interactions between technologies, practices, and social arrangements. When exploring the relational space between these elements, three core concepts become pertinent: creative labour, automation, and distributed agency. Critiquing “creativity” through these conceptual lenses allows us to re-situate the use of generative AI within discourses of labour in post-industrial capitalism and brings us to a conceptualisation of creativity that privileges neither the human user nor machine algorithm but instead emphasises a relational and distributed form of agency." https://link.springer.com/article/10.1007/s00146-024-01921-3
#AI#GenerativeAI#Water#WaterScarcity#Environment: "Once the AI model is in use, each inference, or response to queries, also requires energy and cooling, and that, too, is thirsty work. Ren and his colleagues estimate that GPT-3 needs to "drink" a 16-ounce bottle of water for roughly every 10-50 responses it makes, and when the model is fielding billions of queries, that adds up.
To get the full measure of water use, Ren looked at both the water directly used in cooling servers in a data center and the water used in the generation of electricity the data centers require.
"When we generate electricity using coal-based power plants, using nuclear power plants, we are actually also consuming a lot of water," he explained, and AI is sharply boosting the energy that data centers need.
Add all that water up and apply it to the projected growth in AI in the next few years and the total quickly reaches a staggering level. Ren concluded that global AI demand could result in as much as 6.6 billion cubic meters, or 8.6 billion cubic yards, of water withdrawal by 2027. To put that in perspective, he compared it to how much water some countries use.
It seems to me that the main problem with #ChatGPT and other #LLMs is context. Each new conversation with them is a clean slate and the longer a conversation goes on the slower and more confused they seem to get. I presume taking the context into account means extra processing time, and storage on their part, but moreover they just don't provide a very good interface for communicating with the #AI about a long-lived project. This is critical for #softwareDevelopment.
#AI#GenerativeAI#Research#Science#Chatbots#LLMs: "This new review, led by William Agnew, who studies AI ethics and computer vision at Carnegie Mellon University, cites 13 technical reports or research articles and three commercial products; all of them replace or propose replacing human participants with LLMs in studies on topics including human behavior and psychology, marketing research or AI development. In practice, this would involve study authors posing questions meant for humans to LLMs instead and asking them for their “thoughts” on, or responses to, various prompts.
One preprint, which won a best paper prize at CHI last year, tested whether OpenAI’s earlier LLM GPT-3 could generate humanlike responses in a qualitative study about experiencing video games as art. The scientists asked the LLM to produce responses that could take the place of answers written by humans to questions such as “Did you ever experience a digital game as art? Think of ‘art’ in any way that makes sense to you.” Those responses were then shown to a group of participants, who judged them as more humanlike than those actually written by humans."
Anyone know of any data comparing the energy and water usage of a human being performing a cognitive task - like reading and summarising an article - to genAI doing the same thing?
I suspect the latter would be guesses because these companies are deliberately trying to hide the true costs. But my guess is we will find at least an order of magnitude difference.