Hey folks,
And welcome free readers!
Anthropic raises the feature stakes for chatbots, and SoftBank has an intriguing use for generative models. Meanwhile, Lenovo made me think about getting a tablet from them.
Tom
Big Story
"Anthropic claims its latest model is best-in-class | TechCrunch" https://techcrunch.com/2024/06/20/anthropic-claims-its-latest-model-is-best-in-class/
"Claude 3.5 Sonnet: Anthropic’s AI model is competing with GPT-4o and Gemini 1.5. - The Verge" https://www.theverge.com/2024/6/20/24181961/anthropic-claude-35-sonnet-model-ai-launch
Anthropic released its newest model called Claude 3.5 Sonnet, which it compares (favorably, of course) to OpenAI's GPT 4o. It's supposed to be better at understanding nuanced and complex instructions, as well as being about twice as fast as Claude 3 Opus.
3.5 Sonnet can also interpret charts and graphs with more accuracy and transcribe text from distorted images. Its context window is the same 200,000 as version 3.
Anthropic also released something it calls Artifacts, which lets you create documents and edit them within a workspace inside the chatbot so you can see what it looks like and make edits. Artifact generates it to the side so you can make tweaks. With a text document, you can make edits without having to copy it out to a text editor. With code, you get the UI and can tweak code and see how it affects the way the program runs.
Claude 3.5 Sonnet is available now for free on the web and iOS and on Anthropic's API, Bedrock, and Vertex platforms. Paid users get five times higher rate limits.
The models keep getting more capable but not at the skyrocketing rate they once did, which is to be expected. The Artifacts feature is a good example of how these companies are going to try to get you to use them instead of the competition while pushing the competition to come up with their own better features.
More Stories
"Former OpenAI chief scientist to start a new AI company | Reuters" https://www.reuters.com/technology/artificial-intelligence/former-openai-chief-scientist-start-new-ai-company-2024-06-19/
Ilya Sutskever was a founder of OpenAI and led its safety and security team, which focused on long-term risks from Artificial Generalized Intelligence, something most folks agree is not a near-term risk and there is a debate on how high a risk it will become. Sutskever was also one of the board members who voted to fire Sam Altman as CEO back in November and then changed his mind.
He left the company earlier this month, and OpenAI disbanded his safety team and replaced it with one more focused on near-term risks and dominated by board members.
All of that is a prelude to why it's not terribly surprising that Sutskever has announced his own company called Safe Superintelligence, based in Palo Alto, California, and Tel Aviv, Israel. Safe Superintelligence will also include former OpenAI scientist Daniel Levy and former Apple AI lead Daniel Gross.
The company plans to develop models that will advance capabilities without compromising long-term safety. The company says its business model will shield it from "short-term commercial pressures."
I'm going to guess we don't hear a lot from this company for a while, but it will be interesting to see if and how they get into the race.
"Softbank plans to cancel out angry customer voices using AI | Ars Technica" https://arstechnica.com/information-technology/2024/06/new-emotion-canceling-ai-tech-aims-to-shield-call-workers-from-angry-customers/
Here's an interesting use for generative models that Allison Sheridan from Nosillacast pointed out to me. SoftBank is going to use a generative model to modify the tone of callers to its customer support agents. SoftBank developed the model by hiring Japanese voice actors to express more than 100 phrases with various emotions. The company says the model can recognize anger and hostility and modify it to sound calm and less threatening. It does not change the words, just the tone. Some emotion is still included as well so the support agent can know the customer is upset without feeling harassed.
"Asus Vivobook S15 hands-on review: Snapdragon X Elite toasts the M3 MacBook Air | Tom's Guide" https://www.tomsguide.com/computing/laptops/asus-vivobook-s-15-snapdragon-x-elite-review
"Qualcomm AI/Copilot PCs don't live up to the hype - SemiAccurate" https://semiaccurate.com/2024/06/18/qualcomm-ai-copilot-pcs-dont-live-up-to-the-hype/
I just thought these two stories were interesting side by side. Tom's Guide found the battery life and performance of the Asus VivoBook running on a Snapdragon X Elite chip to be impressive. It says performance beat the MacBook Air. But it also supported the more stridently expressed opinion of SemiAccurate that the Copilot+ software is disappointing. There are a lot of points SemiAccurate makes, but the ones most other reviewers seem to agree on is the AI features aren't ready for regular use and the Arm support of apps is spotty.
"Lenovo squeezed eight speakers into its chonky new tablet - The Verge" https://www.theverge.com/2024/6/20/24182290/lenovo-tab-plus-tablet-android-bluetooth-speaker
The Lenovo 11.5-inch Tab Plus is basically a Bluetooth speaker in a tablet. It has 26 Watts of sound with four tweeters and four woofers. And it can connect to other sound sources to act exactly like a Bluetooth speaker. It is one of those tablets with a kickstand too, which seems essential for this functionality. It also has a 3.5-millimeter headphone jack and supports 24-bit 96kHz high-res audio with Dolby Atmos tuning. Because of all these speakers, it's a little thicker at 7.7 millimeters but still not thick if you ask me. It's available now starting at $289.99.
This is cheaper than some high-end Bluetooth speakers and you get a tablet... I'm tempted.
For Context
"Instagram Recommends Sexual Videos to Accounts for 13-Year-Olds, Tests Show - WSJ" https://www.wsj.com/tech/instagram-recommends-sexual-videos-to-accounts-for-13-year-olds-tests-show-b6123c65
Instagram claims the way the Wall Street Journal got this to happen wouldn't occur in real life. However, it's compelling that they tried the same thing with TikTok and Snapchat and did not get adult content.
"Proton can now back up photos and videos on your iPhone" https://www.engadget.com/proton-can-now-back-up-photos-and-videos-on-your-iphone-with-100036501.html
Keep an eye on Proton. They are really making a compelling case for mainstream users to adopt them and get the extra privacy and security they offer. End-to-end encrypted photo storage from iOS is a mainstream feature.
"Beware — that new VPN you've found could be infected with malware | TechRadar" https://www.techradar.com/pro/security/beware-that-new-vpn-youve-found-could-be-infected-with-malware
I find it notable that for once we're hearing about malware targeting Chinese users. This is how cyberwars look.
"Kenya closes its probe of Worldcoin, opening the door to a relaunch of its orbs after a year-long suspension | TechCrunch" https://techcrunch.com/2024/06/20/worldcoin-in-kenya-2/
Remember how Kenya stopped people from using WorldCoin? Well, they investigated and found no reason to stop it. Remember WorldCoin?
"Microsoft Says Xbox Is Going Big at Gamescom 2024, While Sony and Nintendo Skip the Show - IGN" https://www.ign.com/articles/microsoft-says-xbox-is-going-big-at-gamescom-2024-while-sony-and-nintendo-skip-the-show
"Xbox Will Have its 'Biggest Booth Ever' at Gamescom 2024" https://www.thurrott.com/games/304273/xbox-will-have-its-biggest-booth-ever-at-gamescom-2024
Gamescom in Germany is one of the many events considered to be picking up the slack or preventing the need for E3. But Sony and Nintendo have pulled out of that one too now. Microsoft, on the other hand, is going big at Gamescom with its "biggest booth ever."
Interesting Reads
"For Apple’s AI Push, China Is a Missing Piece - WSJ" https://www.wsj.com/tech/ai/for-apples-ai-push-china-is-a-missing-piece-7d2b0ec6?st=vlewtrmejtak0pk&reflink=desktopwebshare_permalink
"EU cancels vote on child sexual abuse law amid encryption concerns – POLITICO" https://www.politico.eu/article/eu-council-cancels-vote-on-encryption-breaking-child-sexual-abuse-law/
"I Will Piledrive You If You Mention AI Again — Ludicity" https://ludic.mataroa.blog/blog/i-will-fucking-piledrive-you-if-you-mention-ai-again/
"From Infocom to 80 Days: An oral history of text games and interactive fiction | Ars Technica" https://arstechnica.com/gaming/2024/06/from-infocom-to-80-days-an-oral-history-of-text-games-and-interactive-fiction/