Why Humane AI is Not a Total Failure
Happy Thursday everyone,
Welcome free subscribers! If you're interested in what the daily subscribers got, yesterday it was some insight into why China's really easing up on AI gaming. It's not what you might guess. I also wrote up all the Google Cloud Next announcements, mostly about Gemini. And we started off Monday with a report on Apple's change of heart on AI emulators. Of course there's a lot more in there too.
You can upgrade to a yearly paid plan at just $70 for the whole year if you want to get these every day.
For today though we have an overview of the Humane AI pin, that buzzy thing you wear on your shirt that made a great Ted Talk. Also, some details on Apple's repair program expansion and more.
Enjoy!
Tom
Big Story
"Here’s What Reviewers Are Saying About the Humane Pin"
"Humane AI Pin review: the post-smartphone future isn’t here yet - The Verge"
Humane AI is a little voice-activated pin that uses some generative models to answer questions. It projects images on surfaces, like your hand, as well as speaks answers. The device was announced in November for $699 along with a $24-a-month subscription to T-Mobile's unlimited text, talk and data service. The device begins shipping Thursday and reviews have been published. They're not good. Overall reviewers say the device has limited features which work slowly and sometimes not at all.
The way it's supposed to work is you wear it on your shirt, tap the button to ask it something and it does what it can on device and goes to the cloud for the rest. It has a vision sensor to it can use to sense where you are and respond to contextual questions. Humane says it can send texts, take photos, and play music for Tidal.
The Verge's David Pierce describes several of its buggy responses and the fact that it took about 10 seconds to answer a question about the weather and send a text to a friend. It's an ability to recognize items with vision is first on the road map which is good since it doesn't seem to be very accurate at that yet.
Wired's Julian Chokkattu said it identified a temple in Thailandd as being in Cambodia and also got incorrect information about California's high-fructose corn syrup laws.
The user interface felt natural to most reviewers and the projector worked on hands though not in bright sunlight. CENT"s Scott Stein said the on-device translation worked fast though it got stuck in one language once for him.
Also, the battery extender worked well though it tends to get a little hot which is a concern when it's close to your skin. Humane told the Verge that it will add timers and calendars in a software update this summer. So no, it can't set a timer yet.
My first reaction to negative reviews is to look for evidence of piling on or overly high expectations. It looks like reviewers are being pretty fair too Humane AI though. The pin is the right ideas but has horrible execution. That's a positive for me since the idea of wearing a thing on my chest seemed awkward and hard to use. But the objections, with the exception of battery heat, are almost universally about the software. That's easier to fix than the physical product design. If the idea is solid, there's a chance they can get the software up to speed.
My biggest doubt is whether they can stop hallucinations and misidentifications. Those are prevalent in generative models and something I don't use them for. Not using them with the Humane AI takes away a lot of its promised functionality. Definitely don't get one yet. I have one on the way, so you don't have to and I'll pass along my personal review when I get it.
Other Stories
"Apple alerts users in 92 nations to mercenary spyware attacks | TechCrunch"
"Apple drops term 'state-sponsored' attacks from its threat notification policy | Reuters"
Apple sent notifications to an unknown number of users across 92 countries warning them they may the target of spyware attacks. Notably, Apple used the term "mercenary" instead of "state-sponsored" after the Indian government objected to the latter term. Apple sends these kinds of notices to high value targets severs times a year, usually to politicians and journalists.
The real news here is the change in the term. The thinking on the part of India and other governments might be that they want to be able to deny they had knowledge that mercenaries had been employed rather than denying they sponsored the attack. Seems like a fine line to walk, but governments love walking fine lines.
"Apple will open the iPhone to repair with used parts - The Verge"
Apple announced its repair program will allow used parts in self-repairs starting this autumn. Apple requires parts pairing when you repair one of its phones to make sure the serial number of your replacement part matches its database of genuine parts. If it does not match you will get persistent notifications that Apple can't verify the part and for Face ID and Touch ID, they may not work at all. Until now the database only included new parts. Apple says this is to protect the security and reliability of its phones. This expansion means that used displays, batteries and cameras will validate and Face ID and Touch ID sensors will be added in the future. However used parts that have been reported stolen will not validate. Apple told TechCrunch the program will apply to iPhone 15 and later.
"DuckDuckGo launches $9.99 per month privacy bundle with VPN - The Verge"
DuckDuckGo has launched its first paid privacy subscription product. Privacy Pro gives you a WireGuard VPN, personal information removal from data brokers and identity theft restoration for $9.99 a month. All three services will be provided in DuckDuckGo's browser, but the VPN will protect the entire device. The Identity Theft restoration service is provided by Iris, which helps with canceling and replacing documents, freezing credit reports and challenging fraudulent claims.
"Google to launch AI photo eraser for iPhone and other Android phones"
Starting May 15, Google will add several of its photo-editing tools to all users of Google Photos, including iOS users at no charge. These include generative tools reserved previously for Pixel users and Google One subscribers, like Magic Editor, Magic Eraser, Photo Unblur and Portrait Light.
"Adobe (ADBE) Is Buying Videos to Build Its AI Text-to-Video Generator - Bloomberg"
Adobe continues to future-proof its generative models against possible copyright lawsuits. Bloomberg reports Adobe is offering photographers $120 to submit videos of people engaged in everyday activities in order to help train its models. The request asks for more than 100 short clips f people walking, interacting with objects, showing emotions and more. Adobe has trained its models on stock images it owns the rights to and images obtained with permission from creators.
"Taylor Swift Songs Return to TikTok"
In advance of her new album, "Tortured Poet's Department" arriving Friday, multiple Taylor Swift songs have returned to TikTok. Swift's songs disappeared from TikTok along with millions of others owns when TikTok could not agree on a renewal for its licensing deal with Universal Music Group. While Swift's music is distributed by UMG, she owns the masters to her songs and may have negotiated a separate deal.
Neither side is talking either, implying this is an uncomfortable but necessary exception to the two sides posturing in their negotiation battle. And it shows the impressive influence of Swift to be able to work around both large companies to get what she wants.
"New Tech That Asks ‘Are You Sure About Sending a Nude Photo?’ - WSJ"
Instagram will automatically detect and blur nude images sent in direct messages. An on-device algorithm will be used so no data leaves the user's device. If you receive one, you'll see a pop-up explaining how to block the sender or report the chat and telling the recipient not to feel pressure to respond. Senders will get a notice advising them to be cautious about where they send nude photos and telling them how they can unsend them. Instagram is testing the feature in the coming weeks and intends to roll it out globally over the next several months. The feature rich snow is turned on for users younger than 18. Older users will get a notification asking if they want to turn it on.
Context Explained
"Spotify to Let Users Play DJ, Competing With Remixes on TikTok - WSJ"
It's a smart move for the music industry to let companies like Spotify offer these kinds of tools so the monetization can continue to flow.
"Amazon CEO Andy Jassy Touts AI Revolution While Committing to Cost Cuts - WSJ"
This shows Amazon is concerned that customers will not see it as competitive with Microsoft Azure or Google Cloud because it's not a player in AI.
"Huawei building vast chip equipment R&D center in Shanghai - Nikkei Asia"
This is the real test for China's chip strategy. Can it continue to resist US sanction by developing its own tech which it can then sell to Iran, Russia and North Korea and possibly others. Huawei has the chips to do it if anyone does.
Interesting Reads
"European car manufacturer will pilot Sanctuary AI’s humanoid robot | TechCrunch"
"Why Prime Video & Disney+ Exited Original Production In Southeast Asia"
"Is robotics about to have its own ChatGPT moment? | MIT Technology Review"
"Smart rings are meant to be invisible, and that’s the problem"
"China's Q1 GDP growth set to slow to 4.6%, keeps pressure for more stimulus- Reuters poll"