This week, we saw a different side of AI. A side, I knew existed but didn’t want to see.

Last month, Devin (AI Software Engineer) went viral, a few of the folks who got access realised that demo was a trick and the product isn’t really that good. So I took a look at one of their videos (Devin can do any upwork job) and there were a bunch of loose screws.

• Job was cherry picked

• Devin never delivered what customer had asked, but the person responded with “Good job, Devin”

• Some files were fed to lead to response

• The libraries were old even when readme file told exactly what was needed to be done

• Introduced more problems than fixed it.

Covered all that in this short video with demonstration.

I even tried Devika the other day, i was honestly disappointed, you can find that video here.

Next,

Humane AI Pins delivered their first batch and the reviews only talk about the terrible things with the Pin. Some YouTubers called it the worst products they’ve reviewed.

Even top companies like Google have been known to make “smart” demos (remember the initial Gemini demo)?

These companies have a huge hand in making us all AI skeptic.

I’m going to make it a mission to beat the hype moving foward than jumping the boat.

Anyho, let me know what you think? Let’s take a look at our regular content.

Stuff You Should Know

Stable Diffusion 3 API is now live!

Stable Diffusion 3 and Stable Diffusion 3 Turbo are now available on the Stability AI Developer Platform API, as announced on April 17, 2024. These models are built on the Multimodal Diffusion Transformer (MMDiT) architecture, which uses separate sets of weights for image and language representations, improving text understanding and spelling capabilities compared to previous versions.

The Stable Diffusion 3 API is available for non-commercial projects at no cost, while a subscription of $20 per month is required for projects with less than $1 million annual revenue or less than 1 million users.

Generated by SD3:

Google Releases CodeGemma Model 

Code Gemma is a variant of the lightweight, generative artificial intelligence (AI) open model family, Gemma, developed by Google. It is specifically designed for code generation and completion tasks, as well as instruction following12345. CodeGemma is available in three different variants, each with a specific focus on code generation, code chat, and instruction following, or fast code completion on local devices. Google introduces Responsible Generative AI Toolkit & Code Gemma. Toolkit promotes safety & ethics in AI dev. Code Gemma optimized for NVIDIA RTX GPUs, enhancing generative AI on RTX-powered PCs.

Adobe announces potential partnership with SORA!

Adobe Premiere Pro is integrating generative AI video tools from its Firefly family, including third-party integrations with Runway, Pika Labs, and OpenAI's Sora models. These tools will enable users to generate and edit videos using text prompts, extend video clips, and add or remove objects. The integration of these AI models aims to provide users with more creative options and streamline the video editing process within Premiere Pro.

The upcoming features in Adobe Firefly will allow users to make significant changes to video clips through text prompts, such as shifting the mood or style of a scene, generating custom music and sound effects, adding captions, text effects, graphics, logos, and analyzing scripts to enhance the creative process. These AI-powered tools are designed to accelerate production workflows, expand creative possibilities, and ensure the ethical use of content by being based on authorized and legally safe material.

Reka Core:The Multimodal Language Model

Reka AI has introduced Reka Core, a frontier-class multimodal language model comparable to leading models from OpenAI, Anthropic, and Google. Trained on thousands of GPUs, Reka Core excels in multimodal understanding, including images, videos, and audio, with a 128K context window for precise recall. It outperforms models like Claude-3 Opus on multimodal human evaluation and surpasses Gemini Ultra on video tasks while being competitive on language tasks. Reka Core continues to improve, showing promising results in intricate analysis and problem-solving across various industries.

Top Products launched this week

• Gen AI Courses - One AI Platform giving you all you need to learn about Generative AI.

• Vidyard AI Avatars - A whole new way to create personalized videos with Vidyard. Create high-quality AI videos at scale with a hyper-realistic avatar that looks and sounds just like you. Generate personalized video messages in minutes with just a written script.

• Poe 3.0 - Fast AI chat, with access to ChatGPT, Claude 3, GPT-4, and more. Available at poe.com, in the iOS App Store, and on Google Play.

• SpeedLegal - SpeedLegal is an AI tool that helps you understand and negotiate contracts better. It can quickly identify potential risks and explain complicated legal terms in simple language. SpeedLegal also gives you personalized suggestions to improve your contract.

• Collato AI Notetaker- Collato instantly transforms meeting transcripts, images, and audio recordings into documentation. Create an account (it's free), invite [email protected] to your Google Meet, and let Collato do the write-up. Now you can focus on doing more of what you love.

• Reflex- Build a web app in pure Python in minutes. Deploy with a single command. Completely customizable UI. Scale from a small prototype to a full production web app.

Handpicked video

There’s a wearable AI Hype, here’s my contribution:

This week in AI

• OpenAI Japan: OpenAI has expanded to Japan, establishing a Tokyo office to collaborate with local entities and offer a GPT-4 custom model optimized for Japanese. This move supports OpenAI's global strategy, aiming to create safe AI tools that cater to unique needs and benefit humanity. By focusing on Japan's positive image, OpenAI addresses potential concerns and leverages cultural appeal for AI technology.

• Limitless AI’sPendent: Limitless AI has launched the Limitless Pendant, a memory-enhancing AI-wearable that records conversations, transcribes them, and uploads them to a secure cloud for summarizing and drafting emails. The device is priced at $99, with the first units shipping in August. It aims to enhance the user's memory and access to information beyond biological capabilities.

• Google’s Recurrent Gemma: Google's Recurrent Gemma models are lightweight, state-of-the-art open models designed for text generation tasks like question answering and summarization. They offer open weights, pre-trained variants, and instruction-tuned versions, making them versatile and accessible for various applications, including environments with limited resources

• Yann LeCun the God Father of AI Expressed Concerns About the Current State of AI: Yann LeCun, a leading AI figure, criticized AI's current state, stating it "really sucks" and lacks compared to animals' learning abilities. Despite machine learning advancements, LeCun finds it inadequate. He is a professor at NYU and Meta's Chief AI Scientist, focusing on training machines to predict based on videos. LeCun has previously dismissed concerns about AI threatening humanity.

Hope this was a helpful issue. That’s going to be all.