![]() ![]() Fine tuning - optimize small models on CPUs (instead of GPUs) for specific tasks Multiple models - use CPUs for intelligent routingģ. Mixed workloads - combination of online and offline inferenceĢ. ![]() Intel recently gave a talk on data engineering success with Pinterest and shared 3 important use cases for LLMs with CPUs:ġ. Hugging Face has a leaderboard of smaller LLMs - and Intel hit the top of the 7B list on December 1 with its Mistral AI model fine-tuned using Intel Neural Chat.Īnd it even extends to hardware. There's something I don't think most people realize yet: Smaller LLMs can be used for super cost-effective solutions, tailored applications, and rapid prototyping and experimentation (the secret sauce for AI growth!). Smaller language models are making waves in the AI industry too. It’s not all trillion-plus parameter models. I personally see this as the actual (warning: trigger word) metaverse. We are still extremely early in the more "complex" data forms like video, 3D, and actions. For businesses, this means prioritizing investments in technologies that seamlessly integrate various data forms - visual, auditory, textual. Invest in the team and vision, not the niche.Ģ) Some startups share their open roles and upcoming research priorities, some don’t - “openness” is a spectrum and we're going to see companies take a stronger stance at where they fall and some might shift more or less open as a competitive advantage.ģ) Multimodal AI is absolutely the future. Runway ML started with video generation for filmmakers but they have always had an unbelievable research team and creative culture. OpenAI, Microsoft, Meta (tons of M&A in spatial like Scape Tech).ġ) Startups you think are niche players (Runway ML) can land and expand and move into wider plays. You better bet that every company is going after this-basically a multi-modal (including spatial/3D) + physics simulation of the world. Runway is going from LLMs to GWMs, General World Models. Bringing awareness in the classroom to things like this will start to make a difference. MLA and APA both have standards around citing AI generated research. Our focus needs to be on conversations around ethical AI usage and developing a culture around what is and is not acceptable use. I feel like this is a major disservice to teachers, though, who are already benefiting greatly through lesson planning assistance, ideas for differentiation, and support with so many of the administrative tasks that accompany the teaching profession. Many schools are blocking use of AI tools. ![]() It's such unknown territory for educators. There was already an account of a professor at Texas A&M trying to fail his entire class after ChatGPT said that it wrote their papers. Students' academic success shouldn't be tied to an unreliable technology. It concerns me that this is still a feature when they have acknowledged flaws with the technology ( ). Turnitin has a built-in AI detection tool and is a really common plagiarism checker in education. ***original post was incorrectly deleted due to both tech and human error, sorry if you’re seeing it again One more time for the people in the back: DO NOT RELY ON GPT DETECTORS. Say it with me: DO NOT RELY ON GPT DETECTORS. The authors edited an essay sample with the prompt “elevate the provided text by employing literary language”, and it made the average GPT detector performance plummet and go from 70% down to just over 3% □ And catching AI content is just as hard a single prompt can trick the system. □ GPT detectors are biased they misclassified non-English native speaker essays as AI content 61% of the time (compared to English speaker rate of just over 5%) □ GPT detectors as a whole just don’t work this issue is not limited to the OpenAI one (all 7 detectors tested in this paper have issues) We already knew that OpenAI’s AI content detector had a true positive rate of 26% (meaning, when a text sample was actually written by AI, it only caught it 26% of the time).īut now a recent paper out of Stanford University shows that: When OpenAI took down their own GPT detector, I posted on here celebrating it □ ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |