- The Rundown AI
- Posts
- Google's major AI upgrades
Sign Up | Advertise | Tools | AI University
Welcome, AI enthusiasts.
It's been a quiet Summer for Google since May's I/O event — but the tech giant just released an impressive double-feature blockbuster.
With the launch of Gemma 2 and several new Gemini upgrades, is Google about to steal the AI spotlight once again? Let's investigate...
In today’s AI rundown:
Google launches Gemma 2, Gemini upgrades
TIME strikes major deal with OpenAI
Use ChatGPT to turn docs into spreadsheets
OpenAI’s critics train smarter AI systems
6 new AI tools & 4 new AI jobs
More AI & tech news
Read time: 4 minutes
LATEST DEVELOPMENTS
Image source: Google
The Rundown: Google just launched Gemma 2, the next generation of its open lightweight AI model series — alongside new upgrades to its Gemini 1.5 Pro model including a 2M token context window and enhanced coding capabilities.
The details:
Gemma 2 comes in two sizes, a 9B parameter model and a larger 27B parameter model — with a 2.6B parameter lightweight version teased for the future.
The 27B model offers competitive performance to models more than twice its size, while the 9B significantly outperforms similar-class models like Llama 3 8B.
Alongside Gemma 2, Google also opened access to Gemini 1.5 Pro's 2M token context window, allowing for processing of much longer inputs.
New code capabilities for Gemini Pro and Flash enable the models to generate and run Python code to improve accuracy on math and data reasoning tasks.
Why it matters: Google’s Gemma upgrades send a new open-source powerhouse to the top of the rankings while running efficiently (and affordably) on a single GPU. Not to be overshadowed is Gemini’s 2M context window release, which opens up a mind-blowing number of new capabilities to users.
TOGETHER WITH GUIDDE
The Rundown: Transform the way you create and share complex workflows with Guidde, an AI-powered platform that helps users produce engaging video documentation instantly.
With Guidde’s GPT-powered extension, you can:
Design stunning visuals like a pro with no prior experience
Take advantage of AI-generated step-by-step descriptions and voiceovers
Capture workflows effortlessly with Guidde's efficient browser extension
Experience the magic of AI-powered documentation and try Guidde’s free extension today.
OPENAI & TIME
Image source: OpenAI
The Rundown: TIME just announced a new multi-year licensing agreement and partnership with OpenAI, infusing the media giant’s library of content into ChatGPT, AI model training, and other products.
The details:
TIME’s 101-year archive will be used to train OpenAI’s models, with both past and current content also available for more accurate responses to news queries.
ChatGPT and other OpenAI products will surface TIME content in responses, including citations and links back to the articles.
TIME gains access to OpenAI tech to develop new features for readers — with the company already integrating several AI-powered tools on their website.
TIME joins media giants like AP, NewsCorp, and Vox in striking a deal with OpenAI, which comes despite ongoing lawsuits from NYT and others.
Why it matters: ‘If you can’t beat em, join em’ seems to be the theme of the current state of AI and journalism — and it’s getting lonelier on the side of publishers still trying to fight against the current. With the recent news of MTV’s archives being deleted, maybe AI models can not only inform but also act as a preservation tool for years of content currently fading from the Internet.
AI TRAINING
The Rundown: With a simple prompt, ChatGPT can analyze documents, answer questions, perform calculations, and create a downloadable spreadsheet — all in one conversation!
Step-by-step:
Log in to ChatGPT and upload your document to the chat (remember to remove any sensitive data before submitting).
Ask questions about it, e.g., "How much is the security deposit?"
Request calculations, like the total first month's costs, including deposits and fees.
Simple prompt ChatGPT to “create a downloadable budget spreadsheet”, specifying the time frame and desired columns.
Note: Only share documents you're comfortable with. Be cautious with sensitive information and consider redacting critical details before uploading.
PRESENTED BY GALILEO
The Rundown: Galileo Luna is transforming how enterprises evaluate their generative AI systems — offering instant, accurate, and cost-effective detection of hallucinations, prompt attacks, and data leakage.
Luna offers:
Real-time detection of hallucinations, prompt attacks, and data leakage
18% higher accuracy, 11x faster speed, and 97% lower costs than GPT-3.5
No need for ground truth data, saving you time and resources
Customizable evaluation models tailored to your specific needs
Get started with Galileo Luna and experience the future of AI evaluations today.
AI RESEARCH
Image source: OpenAI
The Rundown: OpenAI just published new research exploring how to improve model training using AI ‘critics’ that help human trainers spot errors in AI outputs more effectively.
The details:
OpenAI developed a model called CriticGPT to assist human trainers in evaluating and catching errors in code written by AI systems like ChatGPT.
In tests, CriticGPT caught 85% of inserted bugs in AI-generated code, compared to just 50% for human reviewers alone.
Human trainers also preferred CriticGPT’s code critics over their own 63% of the time on naturally occurring bugs.
OpenAI said that CriticGPT-type models are already being integrated into the company’s training workflow, helping prepare for advancing AI systems.
Why it matters: As OpenAI states in the research paper, both LLMs and AI critics will continue to scale — human intelligence will not. The limitations of our own puny mortal brains look to be no match for evaluating systems that are continuing to soar, meaning AI is quickly becoming its own trainer.
NEW TOOLS & JOBS
🎨 Figma AI - AI built-in Figma, turns text into visual design with ease
📞 Character AI Character Call - Interactive phone conversations with AI characters
💬 Sider ChatScreen - AI chat for on-screen content, available on iOS
❓Question Base - Automatically answer repetitive questions in Slack
🏋️♂️ The Fittest - Gamify your fitness journey with AI
🖍️ AI Gradient Generator - Create gradients from text prompts in seconds
QUICK HITS
A Reddit user shared a clip that reportedly demos early access to OpenAI’s upcoming Voice Mode upgrades, showcasing the ability to weave in background sound effects during responses.
Character AI launched Character Calls, a free feature enabling users to have two-way voice conversations with AI characters across multiple languages on the platform’s mobile app.
University of Tokyo researchers developed a new technique to bind living human skin to robotic faces, potentially enabling more lifelike androids in addition to other medical applications.
Meta is set to begin testing its creator-made AI chatbots on Instagram in the U.S., with a broader rollout expected by August.
Eureka Health introduced Eureka, an ‘AI doctor’ offering personalized care plans, lab orders, and medication adjustments under physician supervision with a specialty in thyroid and diabetes treatments.
New York University researchers published a new paper on MLLMs and vision-centric approaches, also introducing a new Cambrian-1 model family and a new benchmark to measure vision-related tasks.
THAT’S A WRAP
SPONSOR US
Get your product in front of over 600k+ AI enthusiasts
Our newsletter is read by thousands of tech professionals, investors, engineers, managers, and business owners around the world. Get in touch today.
FEEDBACK
How would you rate today's newsletter?Vote below to help us improve the newsletter for you. |
If you have specific feedback or anything interesting you’d like to share, please let us know by replying to this email.
Reply