AI Top-of-Mind for Jan 19
Top-of-mind, and close to home, a thoughtful article in ‘The Mercury News’ titled ‘Fears of Generative AI reminiscent of past moral panics’ on the many Gen AI discussions and parallels to history. Larry Magid’s view: Fears of Generative AI reminiscent of past moral panics. His article also references a good slide deck covering the same.
On models, Sheila Teo in ‘Towards Data Science’ describes how she won Singapore’s first GPT-4 prompt engineering competition. She looks at four areas, most of which I’m not personally familiar with. I have visions of large halls at some future point, much like the computer gaming competitions today.
1. [🔵] Structuring prompts using the CO-STAR framework
2. [🔵] Sectioning prompts using delimiters
3. [🔴] Creating system prompts with LLM guardrails
4. [🔴] Analyzing datasets using only LLMs, without plugins or code —
With a hands-on example of analyzing a real-world Kaggle dataset using GPT-4
And, Hanane Dupouy describes JPMorgan’s ‘DocLLM’ for effectively handling complex documents. Link to the original paper.
On the creative front, another not so positive development, ‘404media’ reports that Google news doesn’t differentiate in ‘quality’ between human and AI generated coverage. So continued proliferation of what the article terms ‘garbage’ and possible lack of real moderation.
Turning to something more positive, a good list of AI-driven video generator tools such as Stable Video Diffusion, Runway ML, Pika Labs, and others. They’ve still got a way to go in accuracy, if my own prompts and resulting clips are evidence. And for TikTok, ‘SocialMediaToday’ looks at the new ‘AI Song’ option, where the platform creates custom music, based on prompts, to accompany clips.
On marketing, if you use Reddit, ‘Practical eCommerce’ offers up some good tools to uncover consumer marketing trends. Tools listed include GummySearch, Marketing Blocks, GigaBrain and Profiler.
Finally, more follow-up from CES, this time an in-depth analysis by Nabil Aloouani in ‘Towards Data Science’ on how Rabbit’s r1 and Large Action Model really operates. The article also takes a swag at a cost analysis given Rabbit’s assumed cloud usage.
