Claude's knockout on GPT-4

Welcome to the 486 new subscribers who joined since last time.

While Elon’s busy suing Sam, Anthropic drops a bomb on us with the launch of Claude 3.

‍

What’s brewing in AI this week:

• Anthropic launches Claude 3
• Elon Musk sues OpenAI
• GPTs: Best new arrivals in the GPT store and I review some featured GPTs
• The other top stories in AI these last two weeks

‍

Dario’s Picks:

I. Anthropic launches Claude 3

‍

‍

Anthropic launched Claude 3 yesterday. Take a good look at the table above – their latest top-tier model, Claude 3, reportedly beats GPT-4 and Gemini Ultra on major benchmarks, including knowledge, reasoning, math and code.

Here’s the details:

• Claude 3 is a family of models and includes (from least to most capable): Haiku, Sonnet and Opus. The models differ in terms of capability, speed, and cost of use.
• The models all have a 200k context window (GPT-4 Turbo has 128k). Reportedly, Opus has a 99% recall ability on long context prompts (ie it can remember stuff really well).
• Anthropic claim’s the models’ vision capabilities to be on the level of GPT-4 Turbo.
• Opus and Sonnet are already available through the Claude.ai chatbot, as well as through the API.

‍

Why it matters: While everyone has been awaiting the ChatGPT vs Gemini showdown, a new heavy-weight has just entered the stage. Leaderboards are one thing, though, I’m excited to see how it stacks up against the others in action.

‍

II. Elon Musk sues OpenAI

Elon Musk is back with harsh critique of OpenAI – this time waging a lawsuit against the company and it’s co-founder Sam Altman. The gist of it is that Elon claims that OpenAI has been transformed into a closed-source subsidiary of Microsoft, thereby betraying its founding agreement to develop AI for the benefit of humanity rather than for profit. He also claimed GPT-4 is already a type of AGI.

Admittedly, I’m leaving out ins and outs here that would be better explained by this excellent YouTube summary: Elon's bombshell lawsuit against OpenAI.

OpenAI’s response, according to an internal memo, is that they reject the claims made.

‍

Why it matters: Some allegations made might be a bit exaggerated, but there seems to be some credible and fully critique-worthy points made by Elon. I’m not sure if the lawsuit will actually go anywhere, but I do think if OpenAI became more open it would be a net win for everyone.

‍

III. Mistral’s new LLM, chatbot and partnership with Microsoft

Paris-based startup Mistral’s been busy lately, they:

• launched a new, cutting-edge model, Mistral Large. Unlike previous Mistral models, it’s not open source. Claims to be the second-best AI (Now 3rd, I guess, with Claude 3 on the market), beating GPT 3.5. However, Gemini Ultra is not mentioned in their comparison. For devs, its roughly 20% cheaper than GPT-4.
• launched a chat interface to their models, Le Chat*. Under the hood, it can use Large, Small or, their prototype model, Next. The Beta is available here. They’re also launching Le Chat Enterprise for businesses.
• partnered with Microsoft on bringing their models to Azure, also making a $16M investment in Mistral. However, the investment has already drawn attention of EU regulators, inquiring into the agreement.

‍

Why it matters: Mistral is just 10-month old but making big waves. While it may not yet be fully on par with the big dogs performance-wise (OpenAI, Google, Anthropic), the launch of Mistral Large shows that smaller companies are catching up fast and creating more options in highly capable foundation models for users and developers.

*if this naming isn’t a nod to Pulp Fiction and the “Le Big Mac” scene, I don’t know anything.

‍

In Focus

AI for companionship - market map by a16z

‍

‍

Justine Moore of a16z is back again with another elegant and useful market map of the AI landscape - this time for AI companionship.

Takeaways about the AI companionship market:

• Companion products live on a SFW-spectrum between 👼 and 😈. It’s ranges from AI for kids and virtual pets, to AI girlfriends and sex bots.
• There’s going to be a place for big companies in both SFW and NSFW markets. Key success factors are likely to be:
• Differentiation versus ChatGPT and regular chatbots (e.g. different UI, less censorship, etc.)
• Interaction modes beyond text (think avatars, personality, animations).
• Memory and progression. Most of use yearn for human relationships that remembers stuff about us and evolve over time.
• Group interactions. Being able to interact with AI companions in virtual or IRL convos with your friends.

Ps. If you’re interested in testing some of the most popular (mostly sfw) companion bots, check out my top lists of character.ai bots and Poe bots.

‍

GPTs

Top new arrivals in the GPT store

Highly rated, new GPTs featured in OpenAI’s official GPT store (from the last week)

1. The Designer's Mood Board   Featured
2. Story  Rank 14 in   Writing
3. Coding Expert  Rank 15 in   Programming
4. Article Writer  Rank 15 in   Writing
5. Mathematical statistics  Rank 14 in   Education

‍

Featured GPTs (sponsored)

‍

FAB (feature advantage benefits) Product Analysis

💬20+

Find easy Feature, Advantage & Benefits of Product, with this FREE AI FAB (Feature / Advantage / Benefits) Product Analysis tool

‍

‍

❞ This GPT works well for getting a concise overview of key product features quickly. In this example I gave it some product information about Vacation Sunscreen’s “Super Spritz SPF 50” and it gave me a concise summary of the features of the product, their related advantages and benefits. Could be helpful for marketers to effectively communicate their product’s value.

‍

God of Prompt

💬1K+

Instantly generates best AI prompts for ChatGPT, Midjourney & Gemini. Choose the AI Tool and describe your idea for a prompt!

‍

‍

❞ I asked it to “Help me prepare for an interview in tech”, and it gave me these four timely questions back. Basically, you give it an idea for a prompt, and it asks you follow-up questions to make the prompt more specific (thus achieving better responses). So you’ll have to put down the effort to answer a handful of questions before getting a “perfect prompt”.. but for important tasks it’s worth it.

‍

SEO & Blog Writer Pro

💬1K+

Expert in SEO Blog Post Writing and Optimization.

‍

‍

❞It features an article wizard that lets you generate an article in 5 steps, allowing you to customise at each step; it also suggests titles and outlines based on SERP. Claims it will give your site an SEO score based on any URL, but when I tried it with whatplugin it replied the site didn’t provide the necessary details. Did a decent jobs on titles and outlines, though.

‍

Expert SEO Article Writer

💬500+  ↑100

Expert SEO article writer for top-ranking, 100% human written articles with Outlines, FAQs and Meta Description.

‍

‍

‍

❞ Creates outlines for blog articles based on any keyword you provide, then writes the entire article for you when you type /continue. It created an outline for a comprehensive article, then proceeded to write it (≈2,000 words). The outline provided a useful starting point and the content seems informative enough, albeit quite dull and lacking originality (as one would expect from a one-click generate article).

‍

Want me to (honestly) review your GPT? Submit it to whatplugin and get featured in the newsletter here.

‍

Bytes

General

• OpenAI launched it’s own community – the new OpenAI forum. The goal is to get domain experts and students together to shape the course of AI. It’s available by invite-only or through applying.
• Google had a lil fiasco with Gemini not generating images of white people. It subsequently paused the image generation of people while working on a fix.
• Apple scrapped their electric vehicle plans, moving their team to focus on AI instead.

‍

Partnerships

• Google Cloud and Stackoverflow are partnering up. The latter is the biggest Q&A forum for developers. Expect AI features in Stackoverflow and better access to coding knowledge in Google’s systems going forward.
• The (previously undisclosed) AI company Reddit did a licensing agreement with ahead of its IPO is Google. And it’s not only Google gaining new training data – Reddit will benefit from Google’s AI to improve its search features.
• ElevenLabs, PrometheanAI and AudioShake are part of Disney’s 2024 Accelerator Program.

‍

Product

• ChatGPT can now read responses aloud to you (works on both web and mobile)
• GPT-4 Turbo’s knowledge cut-off has been updated and is now December 2023.
• Google’s new model Genie, trained on internet videos, can create playable 2D video games.
• Stability AI announced Stable Diffusion 3. Better at focusing on multiple subjects and spelling. The model isn’t yet launched, but waiting lists for an early preview are available.
• Nvidia is on fire lately (well, for a lot of time now), recently crossing $2T in market cap and launching laptop-friendly GPUs for AI. Also check out Wired’s new article with the company’s CEO, Jensen Huang.

‍

Research

• An analysis of 5m freelance jobs to investigate which ones are being replaced by AI.