Tech Insights 2024 Week 36

It has finally happened! We are now at the point where a large language model can generate code that is of such high quality that it is no longer beneficial to write the code yourself, no matter how good your skills are. I have written several apps in Node and Python the past week with Cursor and Claude and it's truly mind boggling how fast you can generate complete applications if you know what you are doing. If you are doing any kind of programming today you owe it to yourself to check out the videos below and download Cursor today!

WANT TO RECEIVE THIS NEWSLETTER AS A WEEKLY EMAIL?

If you prefer to receive this newsletter as a weekly email straight to your inbox, you can sign up at: https://techbyjohan.com/newsletter/ . You will receive one email per week, nothing else, and your contact details will never be shared with any third party.

THIS WEEK'S NEWS:

1. Cursor IDE + Claude 3.5 - A True Revolution for Software Development
2. AI Creates DOOM Video Game in Real-Time
3. Anthropic Launches Artifacts for All Users
4. OpenAI Teams Up with Arizona State University (ASU)
5. ChatGPT Usage has Doubled Since Last Year
6. Alibaba Releases Qwen2-VL : Vision Based AI Outperforming GPT-4o
7. Anthropic Reveals System Prompts for Claude

Cursor IDE + Claude 3.5 - A True Revolution for Software Development

https://www.cursor.com

The News:

• Cursor is a code editor based on Visual Studio Code, and it's' specifically built for AI. Thanks to its integration with Claude 3.5 it has gained a tremendous amount of interest the past week with some truly amazing showcases (see below).
• Andrej Karpathy, previously Director of AI at Tesla wrote: "I'm trying Cursor + Sonnet 3.5 ... over the last few days most of my 'programming' is now writing English (prompting and then reviewing and editing the generated diffs) ... it's a bit like learning to code all over again but I basically can't imagine going back to 'unassisted' coding at this point" - https://x.com/karpathy/status/1827143768459637073

Here are 10 fantastic showcases of Cursor + Claude:

1. A complete video editor with 11k lines of code. 685 commits, 3 full weeks. Cursor, Tailwind, React, Firebase. Web-based, cloud saves, versions, can edit on mobile: https://x.com/MengTo/status/1826955943311241611
2. An 8 year old building a chatbot powered by Cloudflare Workers AI in less than 45 minutes: https://x.com/rickyrobinett/status/1825581674870055189
3. An Airbnb clone created in less than 8 minutes, https://x.com/mckaywrigley/status/1829634304978403706
4. A photo background removal tool created under 20 minutes, https://x.com/mattppal/status/1828829915460075540
5. A native Mac-app to generate voiceovers, created with zero coding skills, https://x.com/ammaar/status/1828131587869024300
6. A Google Chrome extension, created in less than 2 hours "without knowing anything about programming", https://x.com/moritzkremb/status/1828749774499254738
7. An iPhone game with sound, https://x.com/gong_cn/status/1829040895146242185
8. An AI photo generator, created in less than 7 minutes, https://x.com/mckaywrigley/status/1829982464653345178
9. A real-time stock market dashboard, https://x.com/mjovanovictech/status/1829627967250583643
10. An AI version of Trello built with zero coding experience: https://x.com/rileybrown_ai/status/1826734177087316058

My take: Claude 3.5 is exceptionally good at generating high quality software code, but it was missing a development environment that makes it easy to work with the code base in a flexible way. This is Cursor, and the combination is a revolution for programming! If you have not yet picked up AI code generation and you are a programmer, now is the time to make the switch. Like myself, you will probably never look back.

AI Creates DOOM Video Game in Real-Time

https://gamengen.github.io

The News:

• Google and Tel Aviv researchers have developed GameNGen, an AI system that can simulate the classic game DOOM in real-time, running at over 20 frames per second and producing visuals nearly indistinguishable from the original game.
• The AI was trained on 900M frames of gameplay data, resulting in 3-second clips almost indistinguishable from the actual game by playtesters.
• The entire game runs on just a single TPU, and in tests human raters could barely distinguish between short clips of the AI simulation and the actual game.

My take: This is the first time an AI model was able to generate a complex and playable video game in real-time without requiring a game engine. HOWEVER the main reason this works at all is that the original Doom is a "2.5D game" where all graphic elements are drawn as sprites / billboards. The entire game is based on 2D images that are skewed and scaled to fit into a 3D environment. So while it is way too early to draw any kind of conclusions if models like GameNGen could be used for any other type of game, the concept is intriguing and maybe one day we will all play individually adjusted game experiences fully driven by AI engines. For more details see the video below "The future of game development?".

Read more:

• The future of game development... has no game engine? - YouTube

Anthropic Launches Artifacts for All Users

https://www.anthropic.com/news/artifacts

The News:

• 'Artifacts' is a dedicated window in Anthropic Claude AI where you can instantly see, iterate, and build on the work you create with Claude.
• Artifacts mean you do not have to copy/paste code into a separate code editor, but you can instead run it directly in your browser.
• Previously, Artifacts required a subscription, but it is now available for all users, including free accounts.

Five examples of things you can build with Artifacts:

1. Developers?can make?architecture diagrams?from codebases
2. Product managers?can create?interactive prototypes?for rapid feature testing
3. Designers?can build powerful?visualizations?for quick prototyping
4. Marketers?can design?campaign dashboards?with performance metrics
5. Sales teams?can visualize?sales pipelines?with forecasting insights

My take: I love using Claude, and things like Artifacts, Projects and Cursor just make the entire experience better. If you have not yet tried Claude, now would be a good time to give it a try.

OpenAI Teams Up with Arizona State University (ASU)

https://openai.com/index/asu/

The News:

• Arizona State University (ASU) is one of the largest public universities in the United States, serving 181,000 students in a given year and offering over 800 degree options. For nine straight years, U.S. News and World Report has named ASU the most innovative university in America.
• ASU and OpenAI just announced that they will rollout ChatGPT EDU (GPT-4o) into over 200 projects across teaching, research, and operations.
• The university also launched an ‘AI Innovation Challenge’ for faculty and staff, receiving an overwhelming demand for using ChatGPT to maximize teaching, research, and ops.

My take: This is probably the right move to make - to fully embrace large language models and integrating them into the curriculum. With ASU paving the way other institutions like Oxford and Wharton have started to follow, with more coming shortly. Trying to prevent students from using LLMs is probably impossible, so better teach them how to use LLMs the right way.

ChatGPT Usage has Doubled Since Last Year

https://www.axios.com/2024/08/29/openai-chatgpt-200-million-weekly-active-users

The News:

• OpenAI said on Thursday that ChatGPT now has more than 200 million weekly active users — twice as many as it had last November.
• OpenAI also said that 92% of Fortune 500 companies are using its products and that usage of its automated API has doubled since the release of GPT-4o mini in July.

My take: Considering ChatGPT has not improved that much the past year, we can probably expect to see a huge increase in usage the coming year when Project Strawberry / GPT-5 rolls out later this year, which can be used in more domains and than what is currently possible with GPT-4o.

Alibaba Releases Qwen2-VL : Vision Based AI Outperforming GPT-4o

https://qwenlm.github.io/blog/qwen2-vl

The News:

• Alibaba has released Qwen2-VL, a new "Vision-Language" AI model that outperforms GPT-4o in several benchmarks.
• Qwen2-VL can understand images of various resolutions and ratios, as well as videos over 20 minutes long, and excels at complex tasks such as college-level problem-solving, mathematical reasoning, and document analysis.
• Qwen2-VL-2B and Qwen2-VL-7B are released using the Apache 2.0 license, while the Qwen2-VL-72B model is available through API.

My take: If you have the need for a small vision-capable model in your project, Qwen2-VL seems to be a very good fit. Available as open source using the Apache 2.0-license in the 2B and 7B variants, I can see this model becoming widely used in a lots of products in the coming months!

Anthropic Reveals System Prompts for Claude

https://docs.anthropic.com/en/release-notes/system-prompts

The News:

• Anthropic has released the system prompts used for Claude 3.5.
• If you are curious exactly what is sent to the LLM before your text you should definitely check these out.

My take: If you are using Claude on a daily basis you should probably read through the system prompts quickly so you know what the limitations are in the current version. It will save you time and make it easier for you to write high-quality prompts.