ChatGPT plugins and Web Browsing was released in beta yesterday by OpenAI. OpenAI is also preparing to release a new open-source language model, per reporting in that same article. They have also released a tool for identifying which parts of an LLM are responsible for which behaviors. It’s early stages and available in open source on GitHub.
As Google pushes its products into AI, it’s notable that Bard is unavailable in the EU and Canada. It’s in 180 other countries – and the reason is GDPR. And while I’m on Google, the company filed an argument with the U.S. Patent and Trademark Office that AI technology should not be considered an “inventor” by U.S. patent law.
Let’s follow up on the open-source AI model discussion from last week. The Information dug further and cites Ion Stoica, a professor of computer science at the University of California, Berkeley, who helped develop a key open-source AI model using Meta’s technology. Stocia believes Free AI models are now “reasonably close” in performance to proprietary models from Google and ChatGPT creator OpenAI, and most software developers will eventually opt to use the free ones.
Meta announced a new generative AI tool for advertisers called AI Sandbox. It’s being tested with a small group of advertisers and will be rolled out more widely in July.
Anthropic has updated Claude, and the critical update is that the context window now allows for around seventy-five thousand words, or one hundred thousand tokens, up from nine thousand. Compare that to ChatGPT at about three thousand words or the API limited release model of GPT-4 at thirty-two thousand tokens.
Microsoft may have made a splash with the Bing announcement, but users don’t appear to be sticking around. Bing’s global search market share has grown roughly 0.25% since the new Bing launch. As per the Information, the company is exploring bidding on the contract to be the default search engine for Firefox.
Finally, amid all that, I2023-5 It’s AI in medical dictation, and it listens to conversations between doctors and patients and generates medical notes within seconds.
Why do we care?
It’s that last piece I care most about. It’s a practical application, and one can assume it’s one of many. To come will be the various applications and models to apply to customers. In some cases, you’ll be looking for fully formed products; in others, you may be looking for the right model. Now add a suitable licensing model to that, and let’s not sleep on the prospect of open-source models producing better results.