Ever since June of last year with the release of Claude 3.5 Sonnet, Anthropic has been silent when it comes to new models, but that period has come to an end with the release of Claude 3.7 Sonnet, a coding model like no other. Other key takeaways include:
- Microsoft hits the breaks on some US AI data center leases over concerns of an oversupply in infrastructure
- Apple looks to build a giant AI server factory in Texas as part of a $500 billion AI investment plan
- Perplexity opens up a waitlist for people to sign up and try out its own agentic web browser called Comet
Join us at AI Tangle as we untangle this week's happenings in AI!
|
After a long period of model drought and the first hints dropped only a few weeks ago, Anthropic finally reveals what it has been diligently working months on - Claude 3.7 Sonnet. The company touts it as the first "hybrid AI reasoning model on the market," able to bounce back and forth between real-time answers and more "thought-out," complicated answers to questions - and it's already out for regular users and developers alike.
What are the results of Anthropic's months of silence?
The hybrid model, combining the best of both worlds, comes in two modes - one without reasoning and one with, the latter available to subscribers of any of Anthropic's price plans. Most importantly, however, Anthropic claims it has geared Claude 3.7 Sonnet less for science competition problems, and instead shifted focus towards real-world tasks that better reflect how businesses actually use such tech (except for when it benchmarks its models in Pokémon).
For example, on one test to measure real-world coding tasks, SWE-Bench, Claude 3.7 Sonnet won a landslide victory against models ranging from OpenAI's o3-mini-high to DeepSeek's R1 - Claude 3.7 Sonnet was 62.3% accurate at its lowest, o3-mini-high was only 49.3% accurate. With extended reasoning enabled, Claude 3.7 Sonnet also scored more-or-less on par on other benchmarks. However clear that Anthropic is looking to make Claude the go-to model for the software engineering world, debuting a new tool to go along with its code maestro model called Claude Code.
|
Open-Source AI Is Ready for Enterprise - Are You?
"The biggest names in AI are pushing closed models, but businesses need flexibility and independence. I'm co-hosting this session on IBM's Granite to show how open-source LLMs can power real-world enterprise applications."
- Mark Hinkle, your AI Sherpa
|
Microsoft shares dipped on Monday after several analysts reported that the company apparently canceled leases for US data centers with at least two private operators, which could hint at an oversupply of AI infrastructure. According to TD Cowen, "facility/power delays" prompted these cancellations, and Microsoft has scaled back on converting preliminary agreements into full leases. Despite the news, a company spokesperson affirmed that its long-term $80 billion investment plan in AI infrastructure this fiscal year alone remains on track.
In a company announcement made on Monday, Apple unveiled its plans to open a 250,000-square-foot factory in Houston, Texas, together with Foxconn to produce servers for its Apple Intelligence platform. Set to open its doors in 2026, this new factory is part of Apple's $500 billion investment plan in the US over the next four years. It also includes hiring 20,000 new employees focused on R&D, silicon engineering, software development, and AI and machine learning.
Announced via a flashy X/Twitter post, Perplexity is set to launch Comet, its new agentic web browser powered by AI, in a bid to transform the browsing experience. Announced via a flashy X post, the beta version offers little detail beyond a sign-up form for a waitlist, leaving its features and market positioning fairly unknown. As Perplexity expands its AI search and research tools, it faces stiff competition from the usual suspects such as Google Chrome, Firefox, and so on, as well as emerging competitors like The Browser Company’s AI-based Dia showcased back in December.
Much like Google, Microsoft, and Meta weeks before it, Alibaba also announced its own plans to invest at least $52 billion (380 billion yuan) over the next 3 years in its cloud computing and AI infrastructure businesses. The move surpasses Alibaba's total spending in those areas over the past decade and comes shortly after the company released its robust quarterly revenue - Alibaba has been solidifying its leadership position in China's AI race. With its stock up more than 68% this year, Alibaba's AI investment statement is drawing significant investor attention as China's AI market uproar continues.
OpenAI announced in a post on Friday that it is expanding the rollout of Operator, its AI browser agent, to countries including Australia, Brazil, Canada, India, Japan, Singapore, South Korea, and the UK. Initially launched in the US in January, Operator aims to be a jack-of-all-trades agent that can, for example, book tickets, make reservations, file expense reports, and shop online. Currently, Operator is only available to users subscribed to ChatGPT Pro, the company's $200/month plan.
|
Warp - Bring plain English to your command line terminals, allowing developers to accomplish multi-step workflows with AI that's native to the terminal.
Substrata - Substrata analyzes real-time human dynamics to help you sell smarter and close more deals - the better, faster way to close.
Gem - Unify your recruiting tech stack with Gem, a platform powered with AI, CRM, and analytics to provide an all-in-one solution to recruiters' problems.
Pin - Put a Pin on your recruitment processes and let AI that understands your specific needs lend a hand in recruitment sourcing, from first contact to scheduled interviews.
|
Your AI, Your Rules - Deploy Open-Source LLMs at Scale
"Open-source LLMs like IBM’s Granite offer the flexibility and transparency businesses need for real AI innovation. I’m co-hosting this talk to show how enterprises can build AI solutions without giving up control."
- Mark Hinkle, your AI Sherpa
|
Not a Match For Human Creativity (67-min listen)
Joining Decoder podcast host Nilay Patel this time around is Vimeo CEO Philip Mayer in a talk about making Vimeo a different kind of competitor to YouTube and why he's willing to bet that human creativity will beat out AI despite believing that AI is certainly here to stay.
|
| What did you think of this newsletter? Let us know! |
|
|
|
|
|
