Howdy, wizards. It’s one major thing after the next this week.. Google held its I/O developer conference the day after OpenAI’s Spring Update event. Again, there was no shortage of impressive releases. Let’s break it down and also cover the rest of the most important AI news from the last week.
- Google I/O 2024: the key announcements. AI is coming to every corner of Google Workspace.
- Google introduces AI overviews. And content businesses are shaking.
- Former chief scientist Ilya Sutskever leaves OpenAI. Superalignment group co-lead Jan Leike follows suit.
- Claude is finally available in Europe. Europeans now have a viable alternative to ChatGPT and Gemini.
- OpenAI embraces industry standards for AI generated content. DALL-E and Sora are getting digital watermarks.
- GPTs: Highly rated GPTs added to whatplugin in the last week
Dario’s Picks
1. Google I/O 2024: the key announcements
Google announced and demoed a colossal amount of new models, tools and features on their developer conference on Tuesday. If you have the time, sit back and watch the full keynote, or this 10-minute summary.
I’m purposely leaving some things out here for this newsletter not to get too long, but here’s what I consider most important:
- AI overviews in Google Search (covered in detail further down)
- Gemini will be more deeply integrated across Google Workspace. You’ll soon be able to use advanced reasoning and ask questions about your photos and emails. The AI side panel for the rest of the workspace apps (Docs, Sheets, Slides ++) will be rolled out to everyone, with additional features.
- Gemini Advanced, Google’s paid chatbot, is getting file uploads and a data analysis feature – all powered by Gemini 1.5 Pro. Yes, the one with 1m token context window (and it will get upgraded to 2m soon). There’s also a voice assistant coming called Gemini Live, as well as a Gems, a feature similar to custom GPTs in ChatGPT.
- Gemini 1.5, the model, is becoming available to everyone. Not only in Gemini Advanced, but through API and AI Studio as well. Google also launched Gemini Flash – a new model that’s smaller & faster. Considering its performance, it’s very affordable versus other models on the market (it’s on the level of Anthropic’s mid-range model Claude Opus).
- Astra, a new project from Google Deepmind. It’s an AI assistant that talks to you in real-time and understands the world around you through audio and video. Here’s a demo of it.
One thing to note here is that a lot of the new features are not available yet, but will be “later this year”. That means we are still in demo-land, and these products have yet to show that they work as advertised. Google was emphasising a lot in their presentation that “this is real-time”; they have burned their reputation a bit before, for those who remember the scandal with the staged Gemini video from last year.
Why it matters
Google, with so many apps we use every day, is integrating AI deep into every system. Gemini might not be as good as ChatGPT on everything, but Google should not be underestimated as it has unparalleled distribution with its widely used suite of apps and, of course, search (more on that below).
I’d say none of the new launches and announcements by themselves were bigger than what OpenAI announced with GPT-4o on Monday, although they were certainly outnumbered in quantity.
2. Google introduces AI overviews
Google’s SGE (Search Generative Experience) is coming out of its Beta stage and renamed to AI overviews. This means that soon, the first thing you see after a Google search will be a summary of the answer to your query, generated by Gemini.
The reactions to this new tech have been really mixed. Understandably, content publishers have concerns about how what this will mean for their traffic, when users don’t have to click through to their content for the answer anymore. Some have called the new feature a “glorified article spinner” while others think it’s a solid upgrade to the user experience. Yet others, including @levelsio are making the point that traditional SEO (Search Engine Optimization) will shift towards AI SEO (showing up as a source in the summarised AI results).
US is the first country to get AI overviews, with more countries coming soon.
Why it matters
As a user, the introduction of this feature is probably not going to impact you too much – maybe you’ll have to click around a bit less to find what you’re looking for when Googling. For all content businesses out there that have a significant share of search traffic though, this is a huge deal, with some sources I’ve seen estimating 30-60% search traffic drops for most publishers as the feature is rolled out globally.
3. Former chief scientist Ilya Sutskever leaves OpenAI
Ilya Sutskever, co-founder and chief scientist is leaving the company, announced by both himself and Sam Altman via X, as well as in an official press release. He is leaving to pursue, in his own words, “a project that is very personally meaningful to me about which I will share details in due time”.
Hours later, co-leader of the superalignment team Jan Leike also resigned, announcing his departure with a straightforward tweet:
There have been ongoing resignations in OpenAI’s superalignment and safety teams for months, fueling speculation.
Ilya’s role will be taken over by Jakub Pachocki, the company’s current director of research, who reportedly has been instrumental in the development of GPT-4 and related research.
Why it matters
OpenAI’s safety team are jumping ship, and there doesn’t seem to be a consensus about why. Kind of concerning. Some might have seen this coming, though. Ilya was apparently the one leading the failed coup against Sam Altman in November, but later changed his mind. Ever since, his role at the company has been unclear to the public (also, to Elon Musk).
4. Claude becomes available in Europe, and hires Mike Krieger
Anthropic just made Claude, the other leading chatbot on the market after ChatGPT and Gemini, officially available in Europe. This follows the launch of Claude’s API for the European market earlier this year.
They also just hired Instagram co-founder and ex-CTO, Mike Krieger, as their Chief Product Officer. Anthropic will draw on Krieger’s experience in creating new ways for people to interact with Claude, particularly in the workplace, according to CEO Dario Amodei.
Why it matters
Europeans now have a powerful alternative to OpenAI and Google’s chatbots available. Anthropic has very capable LLMs, with their top-tier model Claude Opus currently ranking 5 in the LMSYS leaderboard.
5. OpenAI embraces industry standards for AI generated content
OpenAI is working towards industry standards to identify authenticity of content. They’re joining a widely used standard in the tech industry for digital content certification, C2PA. This is an invisible digital watermark to easily verify the content’s source.
Since the beginning of this year, this verification has been integrated into DALL-E and will soon be added to their upcoming video-model Sora as well.
Why it matters
With the rise of deepfakes and realistic AI avatars, it’s quickly becoming clear to most people that you can no longer trust what you see or hear online. I’m not sure this is the exact standard that will prevail, but standards like this are definitely needed.
GPTs
Top newcomers on whatplugin.ai
Highly rated GPTs that made it into whatplugin.ai’s database in the last week. How rankings work.
'; $(this).html(flexWrapper); }); } else if (index === goodBadColumnIndex || index === pricingColumnIndex) { $(this).addClass('desktop-only'); } }); }); }); $htmlContent.find('table').wrap('
'); $('.blog-post-rich-text').html($htmlContent.html()); $('.blog-post-rich-text a').attr('target', '_blank');});
$(function() { const path = window.location.pathname; const pageTitle = path.split('/').pop(); $('table .tool-name-wrapper').each(function(index) { const anchorId = `#anchor-${index + 1}`; const fullUrl = `https://www.whatplugin.ai//blog/${pageTitle}${anchorId}`; const jumpLink = $('', { href: '#', text: 'Jump to review', class: 'jump-to-review', click: function(e) { e.preventDefault(); const isMobile = window.innerWidth <= 768; const mobileOffset = 50; // Adjust this value for mobile const desktopOffset = 80; // Adjust this value for desktop const offset = isMobile ? mobileOffset : desktopOffset; const targetOffset = $(anchorId).offset().top - offset; window.scrollTo({ top: targetOffset, behavior: 'smooth' }); } }); $(this).after(jumpLink); });});
