- AI Weekly Wrap-Up
- Posts
- 7-19-2023
7-19-2023
Keith Collins MD & Addie Politi
July 19, 2023
Top Story
Meta releases LLaMa 2 model as open source, AND it’s licensed for commercial use (!!!)
Zuck is on a roll. Just 2 weeks after he bitch-slapped Elon released Threads to compete with Twitter, he releases an upgrade to Meta’s flagship LLM. As open source, of course, of course (That’s Meta’s AI brand, and its competitive edge for now). BUT ALSO, LLaMa 2 is licensed for commercial use. For FREE (Unless y’all have over 700 million users). Since LLaMa models tend to be smaller than comparably capable competitors, this opens the possibility of businesses and even consumers running their own chatbots on their own hardware, rather than paying a toll to the cloud-based mega-models (we’re looking at you, OpenAI!)
And for those who want or need to use the cloud, they are partnering with Microsoft’s Azure.
But wait, there’s more:
Meta and cellular chip giant Qualcomm are partnering to bring LLaMa 2 to your very own smartphone in 2024. As in, running natively on your own phone’s hardware, not tethered to the cloud. Meta is serious about getting its LLM into consumers’ hands.
Clash of the Titans
Eight months after ChatGPT, where the hell is Amazon?
Of all the major players, Amazon has seemed to have been caught the most flat-footed. Both Google and Meta have come back strong (see our top story for Meta’s latest coup).
From Amazon? Crickets.
Amazon is the dominant cloud provider, and its rival in this sector, Microsoft, has been cleaning up in the AI space. Despite its seeming lack of movement, Amazon recognizes this threat to its core cloud business, and is reorganizing internally, with a focus on developers. Smart - when you have no product, build an environment friendly to developers, and let them create apps within your ecosystem. Execution will be key, and that has traditionally been an Amazon strength.
Perhaps their biggest asset? Jeff Bezos hates to lose.
Google invented modern AI - here’s how it fumbled the ball
Google researchers kicked off the LLM revolution with their seminal “Attention Is All You Need” paper in 2017. OpenAI embraced the new architecture, and 5 years later unleashed ChatGPT into a shocked world. Meanwhile, Google just sat on the opportunity its own researchers had created, and in the end every one of that storied team left for greener pastures. Complacency and corporate bloat led Google to squander its chance to lead the next tech revolution, and now it plays catch-up in a field that it invented. Sic transit gloria.
Microsoft reveals that its AI “Copilot” will cost $30/month
Most users of Microsoft’s Office suite of productivity tools (Word, Excel, PowerPoint, etc.) pay between $12 and $53 monthly. Microsoft has just revealed that on top of that monthly licensing fee for the base applications, it will charge $30 per month for the AI capabilities for the suite. This could double or even triple the cost of Office for some users. Microsoft, of course, says that the new capabilities will be tooootally worth it. We’ll see if businesses and consumers agree.
Apple is credibly reported to be testing its own ChatGPT competitor “Apple GPT”
Apple has been sleeping on the AI market and something woke them up! Was ChatGPT getting too big? Apple is now running in the AI races with Ajax (Apple-JAX); a foundation to create LLM’s.
Fun News
Israeli agri-tech startups are using drones to harvest fruit
Israel is a hotbed of agri-tech startups. Tevel, one such startup, manufactures smart, autonomous drones that use AI to decide which fruits are ripe enough, then to target, pluck, and place them in receiving containers without damage. Click the link for a fun video of flying robots at work.
Hope, fear, and AI: Verge poll on how people actually use LLMs
Tech blog The Verge polled 2,000 US adults on their opinions about AI. Turns out, 8 months after ChatGPT burst upon the public consciousness, only 57% of US adults have even heard of it. But the consensus is that AI is a Very Big Deal, and people are excited, worried, or both.
ChatGPT is killing Stack Overflow - that's bad for future models
Researchers have found that ChatGPT is significantly decreasing the number of weekly posts to Stack Overflow, the premiere question-and-answer website for programmers. This raises the possibility of an ever-diminishing pool of human-generated content on the internet. This will tend to degrade the performance of future models, which may be trained on data which was largely AI-generated.
AI tool for generating malware now on sale on the dark web
A black-hat LLM known as WormGPT has been sighted for sale in the dark web lairs frequented by cyber-criminals. The model is apparently optimized for generating scam emails, phishing attacks, and other illegal activities at superhuman scale.
LangChain Unveils LLM app debugger LangSmith
LangChain, the premiere open source LLM app development environment, now announces the next tool in the process of getting from demo to ship-worthy product. LangSmith takes your buggy, unreliable demo project, and gives you the tools to test, debug, and otherwise make it ready for market.
Regulation and Politics
FTC reported to be looking into OpenAI for "reputational harm" caused by ChatGPT
Hard-driving chair of the Federal Trade Commission, Lina Khan, hates her some big tech companies is anxious to bring accountability to the leviathans of big tech, and is not afraid to investigate them or sue them in court. The latest target of her wrath zeal is OpenAI, who has been accused of allowing ChatGPT “hallucinations” to defame stiff-necked soreheads innocent people of various misdeeds that have no basis in fact. Khan’s record in court, her ultimate enforcement tool, is mixed, with some of her more aggressive claims being thrown out. OpenAI already has lots of commercial incentives to tame its hallucination problem, and this enforcement action seems unlikely to pose an existential threat to the world’s most famous chatbot.
Google, OpenAI and academics propose global AI governance
A new white paper from Google Deepmind, in collaboration with OpenAI and researchers from leading academic institutions such as Oxford, Harvard, Stanford, Columbia, and others, proposes four potential models for international institutions to govern AI globally.
Go here for an analysis of the white paper:
The original arXiv paper is here: International Institutions for Advanced AI
DEALS
HuggingFace, the “GitHub of AI”, raises $200 million at a $4 billion valuation
The leading repository for open source AI projects is now one of the most valuable AI companies on the planet.
Education
Schools Fight against AI tech invading education
Schools were knocked off their feet when ChatGPT first came out, but now they are ready and armed for the next school year. The AI Education Project is the cavalry for teachers.
Don’t Fear, Robots are here! Will Robots fight against another school shooter?
Santa Fe and other school districts are adopting wage-free AI robot surveillance to keep your kids safe. Would you fear a “5-foot-10, 400-pound robot” around campus?
Hot Take: Don’t Ban AI, use it to your advantage.
New York K-12 teachers are learning about ChatGPT’s grading tools and saying it’s better than theirs. Why fight it? Public schools disagree and are blocking AI from their web servers. Who will win the battle? The AI-enablers or AI-disablers?
That's a wrap! More news next week.