- Everyday AI
- Posts
- Claude 3.5 Sonnet Updates: AI can use computers now?
Claude 3.5 Sonnet Updates: AI can use computers now?
Apple Intelligence beta is live, Google and Qualcomm partner for AI-powered cars, Midjourney's web image editing tool and more!
š Subscribe Here | š£ Hire Us To Speak | š¤ Partner with Us | š¤ Grow with GenAI
Outsmart The Future
Today in Everyday AI
9 minute read
š Daily Podcast Episode: Anthropic just released its Claude 3.5 update and all we can say is daaangg. We show you whatās new and why this will change the future of work. Give it a listen.
šµļøāāļø Fresh Finds: Google makes AI text watermark available, OpenAI adds first Chief Economist and Character.AI & Google faces a lawsuit. Read on for Fresh Finds.
š Byte Sized Daily AI News: Apple Intelligence beta is live, Google and Qualcomm partner for AI-powered cars and Midjourney unveils web image editing tool. For that and more, read on for Byte Sized News.
š AI In 5: Weāre showing you a secret ChatGPT hack to read images inside PDFs. See it here
š§ Learn & Leveraging AI: Anthropic Claude 3.5 comes with some new amazing features. We break down everything you need to know about this update and how itāll change the way you work. Keep reading for that!
ā©ļø Donāt miss out: Did you miss our last newsletter? We talked about the Claude 3.5 update, Advanced Voice mode lands in EU, OpenAI and Microsoft partner for local news. Check it here!
Claude 3.5 Sonnet Updates - AI can use computers now? š§āš»ļø
AI can use computers now?
Yup.
With Claude 3.5 Sonnet updates, Anthropic's LLM now has access to 'Computer Use.'
Is this new mode going to change how we use LLMs? And what else is noteworthy with Claude's new updates in 3.5?
Join the conversation and ask Jordan questions on Anthropic Claude here.
Also on the pod today:
ā¢ Claude Model Benchmarks š
ā¢ New Computer Use Feature š»
ā¢ Potential for Business Applications š¢
Itāll be worth your 52 minutes:
Listen on our site:
Subscribe and listen on your favorite podcast platform
Listen on:
Hereās our favorite AI finds from across the web:
New AI Tool Spotlight ā Topview creates marketing videos with GPT-4o & AI avatars, AMA is an AI marketing assistant and Beloga is a personal AI knowledge amplifier.
Google ā Google is making SynthID Text, its technology that lets developers watermark and detect text generated by generative AI models, generally available.
OpenAI ā Dr. Ronnie Chatterji has been named OpenAIās first Chief Economist.
NVIDIA - NVIDIAās CEO says that the design flaw that Blackwell AI chips had are now fixed.
He also spoke on the EU lagging behind the U.S. and China in AI investments.
Trending in AI ā A new lawsuit is blaming Character.AI and Google for the death of a 14-year-old boy.
Microsoft ā Microsoft Photos is getting a new AI super resolution feature that lets you upscale low-quality photos.
AI Governance - The U.S. AI Safety Institute is at risk of being dismantled if Congress doesnāt authorize it.
AI Models ā Ideogram has launched infinite Canvas for manipulating and combining generated images.
AI in Media ā Over 11,500 creative professionals have signed an open letter demanding the prohibition of using human-created art for AI training without permission.
Read This ā Liquid AI is redesigning the neural network.
AI Security - Researchers have found a 'Deceptive Delight' method to jailbreak AI models.
1. Google and Qualcomm Team Up for AI-Powered Cars š
Qualcomm and Google are set to transform your driving experience with their upcoming "digital cockpit," integrating Qualcomm's Snapdragon technology with Google's Android Automotive OS. This partnership aims to introduce features like intuitive voice assistants and real-time updates that could make your car not just a vehicle, but an extension of your digital life, allowing seamless connections with your devices and even autonomous parking capabilities.
With Mercedes Benz and Li Auto as initial partners, drivers can expect an enhanced journey that anticipates needs, such as automatically finding parking while they enjoy their dinner plans.
2. Apple Unveils AI Image Editing with Watermarks š¼ļø
In a bid to preserve the authenticity of photography, Apple has introduced a new "Clean Up" feature in iOS 18.1 that allows users to remove unwanted objects from images while clearly marking them as modified. Craig Federighi, Appleās software chief, emphasized the importance of maintaining trust in photographic content amidst rising concerns about AI's potential for deception.
Unlike competitors like Google and Samsung, which enable more extensive AI enhancements, Apple is taking a cautious approach, focusing on subtle edits that donāt alter the fundamental meaning of an image.
3. RunwayML Unveils Act-One: A Game Changer for Facial Animation š±
RunwayML has just launched Act-One, a groundbreaking AI model that revolutionizes facial animation by enabling the transfer of an actor's performance directly to animated characters using only video and voice recordings. This innovative technology allows for the creation of realistic animations without complex equipment, requiring just a smartphone to capture subtle details.
Beyond enhancing animated films and games, Act-One also facilitates the portrayal of multiple characters by a single actor in a single scene, opening new avenues for storytelling
4. Midjourney Unveils New Web Image Editing Tool š ļø
Midjourney is set to roll out an enhanced web tool next week, allowing users to edit uploaded images with its generative AI, including a feature to retexture objects based on captions. This move comes amid rising concerns over AI-edited images, as platforms grapple with how to label content generated or modified by AI.
In a bid to prevent misuse, the tool will initially be available to a select group of users, backed by increased human moderation and advanced AI oversight.
5. Canva Unveils Dream Lab with Leonardo.AI š
Canva has just launched its new Dream Lab, powered by Leonardo.AI, which the company acquired three months ago. This innovative hub enhances the platform's capabilities by allowing users to generate stunning visuals in over 15 styles, including 3D renders and illustrations, significantly boosting design possibilities.
With an expanded content library featuring Artlist's Premium Video Library and additional photos from Pocstock, Canva is set to redefine graphic design for creators everywhere.
6. Stability AI Unveils Stable Diffusion 3.5 Update š¤Æ
Stability AI has just launched Stable Diffusion 3.5, a significant upgrade aimed at reclaiming its edge in the competitive text-to-image generative AI landscape. This latest iteration introduces multiple customizable models, including an 8 billion parameter version promising superior quality and prompt adherence, as well as a faster, distilled variant.
Notably, the update incorporates advanced techniques like Query-Key Normalization and enhancements to the MMDiT-X architecture, boosting both image quality and multi-resolution capabilities.
7. Apple Unveils Exciting AI Features Ahead of iOS 18.1 Release š
Apple has officially announced a beta version of its new Apple Intelligence features, including the highly anticipated integration with ChatGPT, which is set to launch publicly next week alongside iOS 18.1. This rollout aims to enhance user experience on newer devices, with innovative tools like Genmoji for creating custom emojis, Image Playground for AI-generated images, and Image Wand for effortlessly removing distractions from photos.
Notably, Siri will now seek ChatGPTās assistance for more complex inquiries, allowing users to tap into advanced AI insights without needing an OpenAI account.
Secret ChatGPT trick to read images inside of PDFs
Can ChatGPT analyze images within a PDF?
Most AI experts would say no.
BUT we found a secret hack to make it possible! We show you how it works.
Check out today's AI in 5.
š¦¾How You Can Leverage:
Anthropic has entered the (AI) chat.
While many people are eyeing Anthropicās updates to Sonnet 3.5 and the forthcoming Sonnet 3.5 updates, we didnāt pay them TOO much attention.
Because we think two simple words could change how we interact with technology in the future.
Computer Use.
What is it?
Put simply, itās talking to Anthropicās updated LLM, in natural language, and then the āComputer Useā mode ā¦. Uses a computer.
Yes. An AI can navigate on a virtual desktop, launch programs, type, click and executive actions.
Just like a human.
Thatās the new LLM-powered feature that Anthropic just revealed, and we gave the new model updates AND the Computer Use mode a deeper dive on todayās show.
So, whatās worth paying attention to and whatās just fluff and hype?
Glad you (rhetorically) asked.
Letās break down key insights from todayās show.
3, 2, 111111111ā¦ā¦
1 ā New model, who dis? š¤©
Alright this oneās a bit confusing.
But a highlight of Anthropicās recent release was updating two of its marquee models.
In short, Claude 3 launched with three varieties:
March 2024
Haiku 3 ā the smallest and least capable, but fast and cheap (via the API)
Sonnet 3 ā the middle of the pack
Opus 3 ā the most powerful and most expensive model to use (via the API)
June 2024
Sonnet 3.5 ā Only the middle model got a shiny update.
October 2024
Connect 3.5 (New) ā Not sure whatās wrong with 3.6, but 3.5 New is available now for the front-end and API.
Haiku 3.5 ā Haiku gets an update, but not yet available.
Opusā¦ā¦. Poor Opus. Still rocking V3
What it means:
OK, model soup. We know.
More on the performance and benchmark below. While we didnāt focus TOO much today on these new updates and performance, we DID do a quick/live rundown on our YouTube channel yesterday.
2 ā To o1, or not to o1? š¤
Alrightā¦. Hereās the benchmarks for us dorks!
With the new updates to Sonnet 3.5 (new) and Haiku 3.5, Anthropic also dropped the prerequisite benchmark chart.
Oh, whatās that at the bottom?
Anthropic straight up said, āNa, weāre not gonna compare ourselves to OpenAIās o1 reasoning model.
* Our evaluation tables exclude OpenAl's o1 model family as they depend on extensive pre-response computation time, unlike typical models. This fundamental difference makes performance comparisons difficult.
Sure, thatās fair.
Buuuuuuuttttā¦.. while Anthropic on one hand says itās not gonna compare itself to OpenAIās Strawberry o1 model, it then cherry picks an instance to ā¦. YES! Compare itself to o1 when itās convenient?
Huh.
What it means:
We went into MUCH greater depth on this in todayās show, including some under-the-hood changes in how Sonnet 3.5 seems to be using a bit of reasoning/Chain of Thought just likeā¦ OpenAIās o1.
But if you want an apples-to-apples comparison of how o1 and Sonnet 3.5 (new) REALLY stack up against each other, here ya go.
3 ā Computer Use: Will we all use it? š§āš»ļø
Alright, now we can get to the main event?
Bet.
Like we said, the new āComputer Useā module is now available in beta, and is used via the API.
(Also in Amazonās and Googleās platforms.)
So while we hope this makes its way to a downloadable version for the masses, right now itās a bit developer-centric.
Also, we gotta give huge props to Anthropic.
Even though the āComputer Useā is far from perfect, Anthropic just shipped it.
No hype.
No waitlists.
Just shipped it.
And admitted that it was flawed and error-prone.
Regardless of your thoughts on that GTM strategy, weād say this approach is a breath of fresh-ish air.
What it means:
Hot dang yāall.
While the beta of Computer Use is super buggy and incomplete while running on a slimmed-down virtual machine and is chock full of guardrails, it does show huge promise.
The fact that today you (and your company) can literally give a LLM natural text commands and it can autonomously use a computer is kinda bonkers.
Transfer data between a PDF and spreadsheet? Yup.
Help research and plan your next business trip on its own? Sure.
Automagically do most of your computer-based work with little human oversight? Not yet.
Hereās the hot take: Computer Use is very beta.
And while the floor is pretty low, the ceiling is CRAZY high on this one.
What are your thoughts?
Now This ā¦
Let us know your thoughts!
Vote to see live results
How do you currently search for things? |
Reply