Hey Devs!

Welcome to the AI for Developers Newsletter—where the AI buzz is hotter than your CPU after an all-night coding session! 

As usual, the pace of change in AI has been nothing short of blistering, and in the few short days since our last edition, the AI landscape has shifted more than a Midjourney image after a style reference code! 

So, let’s dive in and ‘feel the ASI!’ ;) 

Just when you thought OpenAI was firmly back on the AI model throne, Anthropic has crashed the coronation party with the release of Claude 3.5 Sonnet. 

In a move that's part impressive and part "did they even sleep?", Anthropic's latest brainchild is flexing hard — outperforming GPT-4o in several key areas, from coding to reasoning.

Key benchmarks:

Key flexes:

• Sonnet outperforms competitor models on graduate-level reasoning (GPQA), undergrad knowledge (MMLU), and coding (HumanEval).
• It’s now twice as fast as Claude 3 Opus (because patience is so last season).
• It solved 64% of problems in an internal coding evaluation (vs. Opus's 38%).
• It has state-of-the-art vision capabilities. So, it can probably see your future!

Speaking of vision, Sonnet is now acing visual reasoning tasks and deciphering charts and graphs like a caffeinated data scientist. It can also read text from blurry images. 

So next time you can't find your glasses, ask Claude! ;) 

Sonnet’s vision benchmarks

But wait, there's more! Anthropic has also introduced "Artifacts" – a new way to interact with Claude that turns your chat into a “dynamic workspace.” It's like having a smart but slightly showy collaborator constantly working alongside you. Albeit one who never sleeps! 

Even at this early stage, this is an exciting glimpse into the kind of natural language coding future we‘ve heard about from Nvidia’s Jensen Huang.

But, before we crown Claude the new AI overlord, don’t forget that GPT-4 still holds some key multimodal cards. At this point, the AI model race is starting to resemble an ever-escalating high-stakes poker game, only one where the chips are made of silicon.

So, is Claude 3.5 Sonnet the new king of AI hill? Maybe for this week, but as we’ve already seen, next week’s AI monarch will likely be someone new. In the meantime, stay tuned for our deep dive — same AI time, same AI channel!

2024's AI Carnival: Smaller, Smarter, and Surprisingly Powerful

The 2024 AI extravaganza is in full swing, and as we near the halfway point, the highlights are coming thick and fast. Smaller is beautiful. Slimmed-down models are flexing mighty muscles,  AI assistants are more capable than ever, and AI is stirring enough ethical debates to make a philosophy major's head spin! 

Watch in wonder as virtual agents try to convince you they're your new best friend. Discover the marvels, mishaps, and perhaps, the slightly unsettling progress of 2024's AI spectacular. 

Peek behind the curtain in our full report. 

Lamini's Memory Boost: Teaching AI to Stop Fibbing!

Ever wished your AI could remember facts as well as your know-it-all coworker? Lamini AI, the enterprise-grade platform for tuning and deploying open-source LLMs, might have cracked that code.

Enter "Lamini Memory Tuning" - it's like flashcards for AI, but way cooler. This nifty trick helps large language models (LLMs) remember facts perfectly. The result? A whopping 90% reduction in AI "hallucinations" (you know, when your AI starts spouting nonsense).

Here's the kicker: Lamini's method has boosted factual accuracy to 95%, nearly double the usual 50% we see with other approaches. This isn’t just about winning trivia nights—it could be a game-changer for apps that need spot-on factual recall, turning your rambling questions into precise database queries.

Check out our latest deep dive to get the full story.

AWS's $230M of Golden Tickets: All You Can Eat Cloud Credits!

Hold onto your everlasting gobstoppers, kids! AWS is going full-on AI Willy Wonka, and the golden tickets are worth more than a Jeff Bezos rocket launch!  $230 million of cloud computing credits are up for grabs for 80 lucky AI startups. 

Alongside mentorship and a chance to mingle with the cool kids at Amazon's Seattle campus, AWS claims this program will "accelerate development" and help startups get to market quicker. 

But is this a dream come true or a sneaky Oompa Loompa trick? While AWS promises a tour of their Seattle candy factory and mentorship sweeter than a Wonka bar, sceptics might see this as a desperate Amazon ploy to catch up to Google and Microsoft in the AI race.

Dive into our full report to uncover the sweet-tasting details and determine if this golden ticket is your passport to AI Candyland. Your future robot, Willy Wonka, will thank you!

Apple's API Bonanza: When One AI Model Isn't Enough

Apple has been generating plenty of AI buzz of late. It began with its AI-heavy WWDC event, which saw the debut of Apple Intelligence and revealed a partnership with OpenAI. Then, the rumor mill started churning when a Wall Street Journal article claimed the company planned to integrate Meta’s chatbot platform into IOS. 

However, the well-sourced Mark Gurman at Bloomberg has squashed this story. Gurman's report quotes an unnamed source, saying that Apple is evaluating partnerships (including Anthropic's Claude and Google's Gemini) but has yet to decide to move forward.

Either way, Cupertino has decided that (when it comes to AI) more is definitely more.

Catch up with our full WWDC coverage and get the latest AI Apple scuttlebutt, rumors and news here. 

Sutskever's Super Leap: AGI? Nah, Straight to Skynet!

Former OpenAI co-founder and AI wunderkind Ilya Sutskever has finally emerged from his brief digital exile, having stealth-quit OpenAI faster than a teen ghosting their Snapchat streaks. 

His new venture, 'Safe SuperIntelligence' (SSI), isn't just aiming for AGI - because that's so 2023. Instead, Illya and his new BFFs are going straight for superintelligence because why crawl when you can teleport?

With offices in Palo Alto and Tel Aviv, SSI promises its first and only product will be a “safe” supercomputer. Well, nothing says “trust us with the fate of humanity”, like a Silicon Valley startup pinky-swearing to keep mum until it’s ninja-deleted the impending doom of a rogue super intelligent AI!

While critics argue we won't even hit AGI, Sutskever's playing 6D chess. Everyone else is stuck on tic-tac-toe, but Ilya's already won Connect Four with the fabric of spacetime. It's a bold move, but it's our boy Ilya. If anyone can pull off this AI Hail Mary, it might be our boy Illya. Stay tuned for nothing, digital utopia, or robot overlords—nobody’s sure yet!