Google & OpenAI failing against open-sourced models: Doc leaked!
Plus: Google accelerates AI testing. Starcoder: Github Copilot’s free alternative.
Hello, Engineering Leaders and AI enthusiasts,
Welcome to the 13th edition of The AI Edge newsletter. Today, we bring you Google’s leaked document and its controversial predictions. Thank you everyone who is reading this.
In today’s edition:
🌐 Leaked Google doc predicts the end of Google and OpenAI
⭐ Starcoder: Github Copilot’s free alternative
🧠 AI turns brain signals into video
🔓 Google releases its GenAI features to 10x more testers
Let’s go
Leaked Google document predicts the end of OpenAI and Google’s AI dominance
Open-source models will dethrone Google and OpenAI in the AI race- says a leaked Google memo.
and from SemiAnalysis shared it on the publication.The document goes on to cite recent AI breakthroughs by individuals and small companies.
It also mentioned how smaller, cheaper open-source language models are quickly catching up and stirring up the competition.
The author of this leaked document says the enterprise approach of retraining language models from scratch isn’t effective.
They expressed concern about how the time and cost-consuming large models from companies like Google and OpenAI put them in a disadvantageous position compared to open-source models.
The sentiment is amplified by a recent report suggesting OpenAI’s $540 million loss in ChatGPT development. According to the report, Sam Altman thinks OpenAI needs a further $100 billion in the coming years to reach AGI.
The leaked Google doc also mentioned how most LLM knowledge isn’t a secret anymore. People continue to leave organizations like Google and OpenAI and, consequently, take all the learnings with them.
Finally, the doc urges decision-makers at Google to collaborate with open-source contributors to lead the AI race. Given the fact that Google has led the open-source dance in the case of Android and Chromium, many experts find it to be valid advice.
Why does this matter?
The leaked Google memo and the reported OpenAI figures don’t paint a bright picture for the enterprises. AI models are becoming smaller, cheaper, and more efficient- thanks to the contributions of individuals and boot-strapped businesses. However, nothing can be said concretely about how the AI landscape unfolds. Even though it seems like an eternity, we are still in the early days of the AI revolution.
StarCoder: A free code-generating model from Hugging Face and ServiceNow
AI Startup Hugging Face and ServiceNow Research have launched StarCoder, a free code-generating AI tool similar to GitHub's Copilot. StarCoder assists developers in generating code based on natural language descriptions and is trained in over 80 programming languages and text from GitHub repositories, including documentation and programming notebooks.
Why does this matter?
With the talks of open-source AI models taking over, StarCoder is yet another blow for large enterprises. It will be interesting to note how developers report working experience with StarCoder.
New AI breakthrough can read brain waves and reconstruct a video out of it.
Researchers have developed a new AI method called CEBRA, which has the ability to translate brain signals into video. The study was published in the scientific journal Nature. CEBRA has potential applications in areas such as neuroscience and brain-computer interfaces.
CEBRA produces consistent and high-performance latent spaces that can be used for decoding. It can be used for hypothesis testing or label-free analysis and provide rapid, high-accuracy decoding of natural videos from the visual cortex.
Why does this matter?
CEBRA can fill the gap in neuroscience by providing a new method for modeling neural dynamics, as the researchers can quickly and accurately decode natural videos from the visual cortex, and it can provide new insights into the brain’s processing, which is crucial for understanding perception and cognition.
Google amps up generative AI capability testing by increasing the tester pool by 10x
Google Workspace took to Twitter to excitedly announce expanding its pool of trusted testers by more than 10X, widening the scope of customer feedback on its new generative AI experiences and improving them
The company introduced its first set of AI-powered features in Docs and Gmail to trusted testers in March in the United States. And it plans to iterate and refine the features from more customer feedback before making them available broadly to consumers, small businesses, enterprises, and educational institutions in more countries and languages.
Why does this matter?
Since Google enjoys a huge user base, the sooner they launch their generative AI features, the harder things might get for Microsoft. Right now, Microsoft is enjoying the first-mover advantage in AI-powered offerings. Their recent figures are clear evidence.
What Else Is Happening
![[video-to-gif output image]](https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F23c42997-e767-48d9-a1c9-83a5035002d4_600x393.gif "[video-to-gif output image]")
🍕 The pizza ad is completely AI-generated. (Link)
🤖 Google announces MaMMUT- a sophisticated vision-encoder text-decoder architecture. (Link)
🤝 Gitlab and Google collaborated to strengthen privacy with AI-powered security features. (Link)
💬 Slack announces SlackGPT. (Link)
💼 Salesforce and Accenture partner to build a generative AI hub. (Link)
🇺🇸 The White House meets AI leaders. (Link)
📜 AI that can translate ancient cuneiform text. (Link)
Trending Tools
Iconify: Streamlines icon creation process, saving time and money. Perfect for developers, designers, or entrepreneurs.
Kaledio: AI-powered second brain for product teams. Collects and organizes ideas and feedback.
AI Domain Ideas: Free ChatGPT-powered domain name ideation tool. Suggests domain names based on project description.
AI PR Services: Free ChatGPT-based press release generator. Create the perfect press release in minutes.
Adpilot: Create brand aligned ads using AI generated imagery, backgrounds and text.
Ghost AI: Productivity Email App using ChatGPT + Gmail API. Includes automation tools like split screens and keyboard shortcuts.
Keyword Catcher: Generates low competition, high traffic keywords for every niche. Analyze search results and get content ideas with AI.
Creator kit: Generate winning video ads with AI using only Store URL. Multimodal AI for text, image, voice and video.
That's all for now!
If you are new to ‘The AI Edge’ newsletter. Subscribe to receive the ‘Ultimate AI tools and ChatGPT Prompt guide’ specifically designed for Engineering Leaders and AI enthusiasts.
Thanks for reading, and see you tomorrow.