Stephen Downes

Comprehensive overview of the state of large language models (LLM). Here's a sample:

• The GPT-4 barrier was comprehensively broken
• Some of those GPT-4 models run on my laptop
• LLM prices crashed, thanks to competition and increased efficiency
• Multimodal vision is common, audio and video are starting to emerge
• Voice and live camera mode are science fiction come to life
• Prompt driven app generation is a commodity already
• Universal access to the best models lasted for just a few short months
• "Agents" still haven't really happened yet

There's more, and all the links point to sections in this one article.

I agree with this assessment: "We're trapped in an outdated mental model of AI - one that sees it as a mere tool for text generation, plagued by hallucinations, struggling with basic reasoning, and 'easily' contained within institutional policies which just need 'tweaking'. This perception persists even as AI has evolved into something far more sophisticated and pervasive."

Nice summary of the flurry of AI announcements to close out the year 2024. Bryan Alexander describes OpenAI's s Twelve Days of Shipmas, Google's swarm of offerings, Microsft's Vision a spinoff from Copilot, Amazon's family of Nova products, and Meta's Videoseal watermarking tool for AI generated video content. 

Glyn Moody writes, "A central theme of Walled Culture the book (free digital versions available) and this blog is that the copyright industry is never satisfied. Now matter how long the term of copyright, publishers and recording companies want more." Thus their latest land grab: the internet's core routers. Writes Moody: "it's a terrible precedent. It means that blocking – and thus censorship – can be applied automatically, possibly without judicial oversight, to some of the most fundamental parts of the Internet's plumbing." Needless to say I think that this would be a bad idea.

"Friends don't let friends use academia-dot-edu," writes Kathleen Fitzpatrick. The same goes for Research Gate. These services are based on collecting academic content and then making it hard to access. What they really want you to do is 'upgrade to premium'. Hence the need for this service: "open access, community governed, and free to all - and permits downloads without requiring a login." I need to figure out how to make it work best with this newsletter.

The Slashdot thread that discusses this story has all the commentary you would need on it. Specifically, the author blames computing for: hate on Twitter, anti-Rohingya content, a youth mental health crisis in the U.S., negative behaviours like bullying and exclusion, and body image issues. As one /. commenter says, the title needs to be fixed: "Blaming Media, ACM Publication Argues Media 'Has Blood On Its Hands'." As someone who was around before social media, I can attest that none of these is unique to the computer age. Mass media proliferated hate and misinformation as efficiently as any computer network. The only difference is that it has been democratized; it is no longer a corporate and state monopoly. As the other articles today show, though, companies and governments are working hard to regain that monopoly. Via Apostolos Koutropoulos.