Google Uses It's Size to Its Advantage
Hey folks,
A LOT about Google I/O. I start with my summary impression of the announcements and then just included my raw notes. Plus there’s lots of non-I/O news as well.
Enjoy!
Tom
Big Story
Google I/O.
Overall Google went to great lengths top show its prowess in AI, emphasizing its large context window as its biggest advantage. It did not do a lot of live demos. Google is not acting like it's behind. Just emphasizing how it can add AI into your life because you're already living on Google through search, Android and Workspace. Google's models may or may not work as well as Anthropic's or OpenAI's but they're good enough. For a lot of developers and consumers they will be the easy choice since they already use Google products. It may not sway people to move to Google though. Google's pitch is similar to developers. You already develop for Google products like Android, stay here and take advantage of our AI without having to work hard to integrate it. Google's using its ecosystem to its advantage.
And like OpenAI on Monday, Google is really showing off the ability for you to just talk in at the way you would normally talk to an AI agent to get the answer you need. The ability to point your camera at something and ask a question about it sounds impressive. Especially when troubleshooting things like broken dishwashers and such. Trip planning was another big example that I think looks great in demos but probably won't be all that useful for most people all that often.
The question I really have is how Search is going to work better but also generate revenue for Google without undermining the Web. We got a partial answer on how it will deliver its search result better. We heard almost nothing about the rest. This is a developer's conference so maybe that is bets left for other gatherings. But I was surprised to hear no acknowledgement of that issue. The improvement of search is a big advantage for Google since everyone uses them and they have the biggest factual database of the Web in the world. Given that it seemed slightly underwhelming how Search AI was presented. Most fdo the feature felt like slightly faster or well-organised responses. The video search was the big impressive addition.
Here's my notes as the announcements happened.
- AI Overviews launching in search to all US users this week. This is what used to be called Search Generative Experience.
- Ask Photos helps find specific images with natural language.
- Gemini 1.5 Pro improved version available to all developers globally. 1 million token context window. Also available to consumer sin Gemini Advanced. An expanded 2 million context window is available in private preview to developers.
- Gemini 1.5 Pro available no win Workspace Labs.
NotebookLM.
- Gets Gemini 1.5 Pro
- You can take a bunch of files and create a personalized audio guide based on those inputs. And you can interrupt and ask questions.
Agents
- Demonstrated showing a picture of shoes and saying you wanted to return them. Agent found receipt and return policy and created a return label for you.
- Updated address cross multiple services
Google DeepMind CEO Demis Hassabis
Gemini 1.5 Flash
- Lighter weight model compared to Pro, fast and cost-efficient, with the same 1-2 million context window.
Project Astra
- Universal agent. Recorded version of the demo OpenAI did live yesterday.
Imagen 3
- more photorealistic and richer details.
- understand prompts the way people write.
- Best yet at rendering text
- Sign up at labs in IMageFX and coming to Vertex AI
Music AI Sandbox
- Generative Music
Veo
- more photorealistic and richer details.
- Veo will be in VideoFX
- Coming to VideoFX in labs
Google Cloud customers
Trillium
- New TPUs 4.7x improvement and will be available in late 2024 to Cloud customers.
Axion Processor
- Custom ARM-based CPU
NVidia Blackwell GPUs will be available in early 2025.
Search
- Customized Gemini model for search
Multi-step reasoning. - find studios near you, available times, and how to book. All in one query.
Google stressing its advantage in having access to facts
Custom workout routine, party planning, meal planning etc.
- "Helpful clusters" organize info into a complex search result. You'll start to see this new layout starting with restaurants.
Ask questions with video
- video frame fed into long contact window to recognize info and create overview
Rolling out in coming weeks. First to search labs opt-ins
Gemini for Workspace
- Gemini side panel will be generally available next month
- Gemini in Meet is expanding to 68 languages.
Gmail Mobile
- Summarize feature for threads
- Q&A on anything in inbox.
- Reply prompts take in context
Labs users. Summarize comes this month; Q&A and Contextual replies arrive in July.
AI Workflows for documents and Gmail. Like extracting receipts from Gmail into an expense-tracking spreadsheet
- AI workflows will come out to lab users this September, along with data Q&A.
Virtual Gemini-powered teammate
- Has a workspace account
- Give it a job role. Like monitor and track projects, Add to chats
- third-parties will eventually be able to create their very own versions of Chip.
- No date.
Gemini App
- Live - interact with voice - coming in summer
- video understanding
- Gems - create a virtual expert. Rolling out in coming months.
Gemini Advanced
Trip planning in Gemini Advanced. - Coming this summer to Gemini Advanced
- Gemini Advanced has the longest context window of any chatbot in the world (eventually 2 million tokens)
- Coming to 35 supported languages now
Android
- Search
Circle to search getting more complex problems later this year.
- Gemini is the AI assistant
Context aware. Can ask questions about a video you are watching
- On-device models for speed and privacy
Coming to Pixel later this year is Gemini Nano with Multimodality.
Talkback - accessibility feature getting multimodal capability.
Gemini can understand phone call and warn about scams.
Later this year can understand content of screen.
Android 15 Beta 2 details coming tomorrow
Gemini 1.5 Pro and Gemini 1.5 Flash Developer announcements
Available globally in more than 200 regions
API features - video frame extraction, parallel function calling, and context caching (saves money by ensuingg repeat duplicate data once).
- $7 for 1.5 Pro, $3.50 up to 128K, Flash is 35 cents for one million tokens. (GPT-4o is $5 per million input tokens and $15 per million output tokens.)
Ships next month
AIstudio.google.com to freely try models
Gemma - Open Models
- First vision language open model PaliGemmaac. Optimized for images
- Gemma 2 - coming in June. 27 billion parameter model optimized by Nvidia and can run on single TPU in VertexAI
AI ethics and responsibility
- Working on tools to prevent misuse of its tools like IMagen3
- SynthID watermarking expanding to text and video.
- Collaborating with the C2PA standards efforts.
- LearnLM is a new family of models based on Gemini fine-tuned for learning.
- Learning Coach Gem launching in the coming months
- LearnLM in YouTube makes educational videos interactive.
More Stories
"Sony PlayStation will soon have two CEOs"
"Sony Has Now Sold 59.3 Million PlayStation 5 Consoles - Thurrott.com"
"Sony 2023 earnings: Tech giant misses trimmed PS5 sales target"
Image-sensor maker Sony saw strong sensor sales to drive a 14% year on year increase in revenue though it was the first quarterly drop since September 2020. Operating profit was up 57% on the year. Financial services dragged down overall profit and revenue leading to a 7% drop for the full year over last year. Video games also didn't help as Sony missed its own lowered target of PS5 sales by a small bit. It came in at 20.8 million, when they had expected 21 million. Sony projects 18 million PS5 sales this quarter.
Sony also announced that Sony Interactive Entertainment, which oversees the gaming business, will get two new CEOs. Jim Ryan announced his departure in March and the president, COO and CFO of Sony Group Corporation has been interim CEO. Hideki Nishino becomes CEO of SIE's Platform business group overseeing hardware and third-party content. Herman Hulst will be CEO of SIE's Studio Business Group overseeing content development, including TV and movie adaptations and first party publishing. They'll both report to Nishino when they take on their new roles starting June 1.
There are a few alarm bells ringing int he gaming division. Sony's not in trouble but it needs a clear approach to the changing nature of things before it has to scramble.
"Exclusive: Google is experimenting with running Chrome OS on Android"
Android Faithful's Mishaal Rahman, writing for Android Authority, reports that Google demonstrated a way to run ChromeOS on an Android device. At a private event, the project code named "ferrochrome" ran in the Android Virtualization Framework on a Pixel 8. No word on when if ever this would become a shipping product. I am very curious what if any use Google puts this to. The popular and reasonable expectation is some kind of Samsung Dec style docking. Could be very cool.
"Comcast to Launch Peacock, Netflix, Apple TV+ Bundle"
The bundles keep on coming. Later this month Comcast will offer its broadband, TV and mobile customers a bundle of Peacock, Netflix and Apple TV+ at a reduced price. Comcast will call the bundle the StreamSaver. The company did not announce a price but implied it would be a deep discount over getting the services individually. The cheapest way to get all three services at this point would cost $23 a month with ads on Peacock and Netflix. Though Peacock is getting a price increase of $2 on its lowest tier starting July 17.
If I may, this is going to keep happening in multiple combinations for the next year or so. Bundles will come. Bundles will go. Do not be surprised. This is intentional. The companies are trying out models to see what works.
Square Enix reported that sales rose 2.6% last fiscal year but profit fell 15.8%. This despite the launch of Final Fantasy 16 and Final Fantasy 7 Rebirth, both PlayStation exclusives. Square Enix announced a three-year plan to “aggressively pursue a multiplatform strategy that includes Nintendo platforms, PlayStation, Xbox, and PCs.” They didn't say this would mean an end to Final Fantasy exclusivity to Sony, but it probably does. They also talked about emphasizing quality over quantity. Looks like Multiplatform is the wave of the gaming future.
"Biden administration quadruples import tariff for Chinese EVs"
"US Sued Over Blacklist on Firms Linked to Chinese Military - Bloomberg"
"Apple supplier Foxconn's first-quarter profit jumps 72% but misses forecasts"
"Alibaba (BABA) earnings Q4 2024"
"Tencent earnings report Q1 2024"
There are a lot of small stories impacting China today, so rather than take up all my "For Context" space with them I'm going to bundle them all together for you here.
The US will increase tariffs on a number of Chines goods, including chips, solar cells, medical products and a quadrupling of tariffs on EVs. Tariffs not only raise the price of the goods they target for import but can raise the price of domestic goods as well as there is less price competition to keep prices low. So it's possible you'll see higher prices for all these kinds of goods.
One Chinese company is fighting the US decision to place on the "entities" list which restricts who can sell it restricted items like chips. Hesai Tech group makes sensors for autonomous cars and is suing to overturn the designation that it supplies parts to the Chinese military. It says it only sells to commercial and consumer clients.
And we got a trio of earnings reports. Foxconn, which is Taiwanese but has several mainland Chinese factories saw a 72% rise in profits but that was still a little lower than expected. But it still expected Q2 revenue to grow significantly. E-commerce giant Alibaba saw an 85% drop in income and an 86% drop in profit. The company is dealing with the slowed demand in China and a slow increase in international business. Entertainment-oriented company Tencent, which makes WeChat, on the other hand, showed its fastest profit growth in three years thanks to online ads and business services. Its gaming sector improved but is still slow.
So overall Chinese companies feeling the pinch from US restrictions and a slow economy at home, but ads and chips are going strong.
"Warner Bros. gives Adult Swim games back to their creators rather than kill them | Ars Technica"
Warner Brothers Discovery has been saying it will shut down about a dozen Adult Swim-related indie games games for a couple of months now but took its time in clarifying that it will do the right thing and return the games rights and store pages to the developers. Initially the company had said it would not transfer the games to their creators but thankfully changed its mind. Developers could have resisted their games but would have to start from scratch on recommendations and reviews.
For Context
This means you can I and everyone can make reactions videos to diss tracks from either artist without being demonetized.
"Amazon-backed Anthropic launches its Claude AI chatbot across Europe"
It takes a little longer for companies to launch chatbots in Europe as they have meet different rules there. But this shows that it's mostly paperwork not substantive changes that need to happen.
The board overturned Facebook's own decision. The reasoning was that while it technically violated the rules, its newsworthiness outweighed that because of public interest in the topic and reporting. The board said the documentary was made to raise awareness, not sensationalize the issue. It's an interesting decision that gruesomeness and offensive imagery are not always grounds for removal. Context is important according to the board.
"US opens probe into Alphabet's Waymo over 'unexpected behavior' of self-driving vehicles | Reuters"
17 reports of minor crashes into stationary objects like poles and 5 reports of possible traffic law violations. I'm thinking there's a chance the NHTSA doesn't finds a problem. But we'll see. First time under the microscope in a while for Waymo too.
"EU says Booking.com must comply with strict tech rules, investigates X | Reuters"
The EU is following its promise to identify new gatekeepers under the Digital Markets Act. Booking.com joins Alphabet, Amazon, Apple, ByteDance, Meta and Microsoft. It passed the threshold of 45 million monthly active users and 10,000+ yearly active business users on March 1. The designation just means it has increased obligations for interoperation and fair play that its competitors do not.
Interesting Reads
"The hunt for rare bitcoin is nearing an end | Ars Technica"
"‘Noise’ in the machine: Human differences in judgment lead to problems for AI"
"The Walls Are Closing in on John Deere’s Tractor Repair Monopoly"
"How the Middle East became a powerful force in AI, tech development - The Washington Post"
"German Companies Bet on AI But Payoff Could Be Years Away - WSJ"