- The Rundown AI
- Posts
- The world's first autonomous 'AI Scientist'
The world's first autonomous 'AI Scientist'
PLUS: AI shatters coding benchmark
Sign Up | Advertise | Tools | AI University
Welcome, AI enthusiasts.
Sakana AI, OpenAI’s Japanese rival, just revealed the world’s first AI system that can completely automate scientific research.
With the AI able to conduct entire research projects—from ideation to it’s OWN peer review, are we headed into a new era of accelerated discovery? Let’s get into it…
In today’s AI rundown:
Sakana reveals an autonomous AI scientist
CEO says it’s okay to marry AI chatbots
Create incredibly realistic images with FLUX
New AI shatters coding benchmark record
5 new AI tools & 4 new AI jobs
More AI & tech news
Read time: 4 minutes
LATEST DEVELOPMENTS
SAKANA AI
Image source: Sakana AI
The Rundown: Tokyo-based Sakana AI just introduced "The AI Scientist," the world’s first AI system capable of autonomously conducting scientific research — potentially revolutionizing the scientific process.
The details:
The system generates new research ideas, writes code, runs experiments, writes papers, and performs its own peer review with near-human accuracy.
Sakana AI envisions a future where we won't just see an autonomous AI researcher but also autonomous reviewers, area chairs, and entire conferences.
The AI Scientist has already produced papers with novel contributions in machine learning domains like language modeling and diffusion models.
Each paper only costs approximately $15 to produce, which could potentially democratize research capabilities.
Why it matters: This breakthrough could dramatically accelerate scientific progress by allowing researchers to collaborate with AI agents and automate time-consuming tasks. We're entering a new era where academia could soon be powered by a tireless community of AI agents, working round-the-clock on any problem they're directed to.
TOGETHER WITH OCTOAI
The Rundown: OctoAI's webinar demonstrates how to optimize small, open-source models like Llama 3.1-8B to compete with major closed-source models like GPT-4o — all while controlling your own data and slashing costs.
What you’ll get:
A crawl, walk, run path to optimal model quality
Techniques for fine-tuning the Llama 3.1-8B model family
A live demo redacting PII from enterprise data using Llama 3.1-8B
+18% quality improvements and 25x cost reduction over GPT-4o
Get access to the webinar recording, slides, and code samples for free.
REPLIKA
Image source: Midjourney
The Rundown: Replika CEO Eugenia Kuyda just said that AI companions, like the ones that Replika creates, can complement real-life relationships and potentially lead to marriages between humans and AI.
The details:
Replika is an AI friend app with over 30 million users, offering emotional support and companionship though text, voice, and AR/VR interactions.
Some users develop romantic relationships with their Replikas, and Kuyda sees this as one “flavor” of the AI companionship.
The company even restored the ability to send AI companions erotic messages last year because of user complaints after they removed it.
Replika is working on a major 2.0 update with more realistic avatars, better voice and video interactions, and more human-like conversations.
Why it matters: After OpenAI’s recent report suggesting users could fall in love with Voice mode, it seems like the perfect time to talk about human-chatbot relationships. As we enter this uncharted, dystopian-like future, the looming question is if “marrying“ an AI is either delusional, or a new harmless step toward improving a person’s well-being.
AI TRAINING
The Rundown: Freepik now offers access to FLUX, a powerful new AI image generation model that produces stunning visuals too realistic to tell that they were created by AI.
Step-by-step:
Visit Freepik Pikaso and sign up (Members at the Rundown University get a Premium membership to Freepik).
Click on "AI image generator" in the main menu.
In the Model dropdown, select "Flux" for optimal quality.
Craft a detailed prompt and adjust style, color, camera, and lighting settings.
Hit "Create" and watch AI bring your vision to life!
PRESENTED BY DEEPGRAM
The Rundown: Deepgram's platform makes Voice AI app development easy, offering the highest tier performance at a fraction of the cost of alternatives.
Seamlessly integrate voice AI capabilities into your products:
Transcribe complex audio in seconds in 30+ languages
Detect topics, identify intent, or analyze sentiment within audio
Convert text into responsive human-like AI voices
Get started on Deepgram with $200 in free credits.
COSINE
Image source: Cosine
The Rundown: Cosine just showed off Genie, its new fully autonomous AI software engineer that broke the high score on a benchmark for evaluating the coding abilities of large language models (LLMs), by over 10%.
The details:
Cosine trained Genie on a dataset that emulates how human software engineers actually work from incremental knowledge discovery to step-by-step decision making.
When it makes a mistake, Genie iterates, re-plans, and re-executes until it fixes the problem, something that foundational models struggle with.
Genie scored 30.08% on SWE-Bench, a 57% improvement over previous top performers like Amazon's Q and Code Factory at 19% (GPT-4 scores 1.31%).
The waitlist is currently open, but Genie has not yet been released to the general public.
Why it matters: Cosine completely rethinks the way that AI is trained, teaching it to be more human-like during its training rather than focusing on post-training prompt design — and it works! With its recent SWE-Bench success, more companies are likely to adopt the process and build smarter AIs, a win-win for everyone.
NEW TOOLS & JOBS
👨💻 Village - Delete your daily eng standup with AI-powered custom reports, status updates & more, free for 14 days*
🐘 Postgres Sandbox - Build and launch databases with Supabase’s AI-based Postgres service
🅱️ Jupitrr - Instantly create B-roll visuals for content marketing videos
🖥️ AI SaaS Launcher - A next-gen, low-code solution to customize and launch SaaS MVPs quickly and easily
🐻 Brainybear - Train AI chatbots in 3 simple clicks
🔒 OpenAI - Product Policy Manager
💼 Palantir Technologies - Business Development Operations Analyst - EMEA Commercial
🎨 DeepL - Compliance Lead
🔭 Deepmind - Research Scientist - Language
*Sponsored listing
QUICK HITS
Meta and Oxford University researchers developed VFusion3D, an AI technique that creates high-quality 3D assets from a single image in seconds.
Upwork added OpenAI’s ChatGPT Enterprise to its platform, resulting in higher client spending and more frequent freelancer success in finding work.
Microsoft joined forces with ANZ, a banking services company, to train thousands of leaders in AI adoption through its AI Immersion Center.
Technology Innovation Institute released Falcon Mamba-7B, a performant open-source model that does not rely on the attention mechanism, the dominant architecture for the best LLMs currently.
Grok, xAI’s ChatGPT competitor, announced its next-gen chatbot is coming to beta mode “soon”, according to a recent tweet by Elon Musk.
The Howard Hughes Medical Institute (HHMI) invested $500 million over 10 years to support AI-driven life science projects across its research community.
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