I write a newsletter about startups and investing—for ai builders of all levels.
I record mini-tutorials, review tools I’m testing, share my insights and give you a peek behind the digital curtain from an exited founder turned investor.
Hey folks,
its glorious weather here in the UK, which is perfect for something i’ve wanted to do…
are you a UK uni student? we want to help you.
thinking of starting a company one day? using/studying AI in uni? want to get involved in startups? meet other like-minded students?
i live in the centre of the ‘golden triangle’ of oxford/cambridge and london. i’ll be looking at some irl initiatives - maybe a hacker cottage for a weekend hackathon?
i spoke at oxford university in 2022, and i want to do more.
i’ve invested in several uni grads (oxford mainly) and i want to find more students looking to start companies one day (and be your first cheque!)
is that you? apply here
p.s. The Create With AI Conference is returning to London on May 22nd, 2025. Use BENBITES for 10% off if you're planning to go. See you there?
🔎 News worth knowing
OpenAI Updates:
Health Bench - a new benchmark that tests how good the models are at flagging emergency situations, doing data work in healthcare, accurate communication and more. o3 does best, grok is #2 with gemini 2.5 pro at almost the same level.
4o-mini gets Reinforcement Fine-Tuning - RFT teaches models to reason in a certain way to increase accuracy on tasks where answers can be clearly verified. This guide from OpenAI has use cases on how companies are using RFT.
Deep Research has a GitHub and Sharepoint integration now. So you can let chatgpt go wild on a repo or your internal docs. And then export those deep research reports as PDFs, making them easier to read, share and archive. (Grok launched PDF creation last week and had everyone raving about it—makes sense oai is hot on their heels)
Bolt.new now supports two way sync with github, plus options like creating a branch for developing new features etc. On my list to check out how smoothly it works.
AssemblyAI just dropped a prompt-based Speech Language Model that boosts transcription accuracy with zero retraining. Now, AssemblyAI offers Speech AI models that transcribe and understand conversations better than ever, making it easy to build voice-powered apps. Start for free and ship faster.*
Google's pushing what Gemini can understand from videos. Gemini models from day 1 are capable of that. Technically, it's not “video” but images (sampled at 1 frame per second) + audio interleaved together. Previously a) the performance was not that good and b) you had to do the work of passing the audio, images, etc. on your own as a developer. Since late 2024 gemini has the ability to accept raw video directly from their SDK and the new Gemini 2.5 models are really good so that 1fps is enough for most tasks.
*sponsored
want to partner with us? Click here
🌐 What I’m consuming
few essays I’m reading on the new internet with AI:
personalization of AI by Bojan Tunguz
when software buys software - how do you sell a tool built to be used by AI
5 biggest problems with today’s conversational chatbpt design by Julie Zhuo
A good vibe check on Gemini 2.5 Pro and Flash by friends at Every.
Thinking about automated news? This piece on building news agents covers how you might go about it. (yes, it includes MCP)
Hassan’s ai apps always go viral. nice peek into his process from ideation to launch.
This fart sound generator on websim, because why not? You can generate all sorts of sounds on websim now. (i’m an investor)
Sakana AI Labs is out with something called Continuous Thought Machines (tweet here). The folks at Sakana keep comping up with these wild ideas. Related Q: should we do a post on all the new “AI labs” founded in last 1 year?
⚙️ Tools I’m tinkering with
Underleaf - a suite of LaTeX and writing tools for your next research paper.
Kit: An open-source toolkit specifically for building AI dev tools.
coupling a bunch of borswer.desktop agent tools →
Scout.new: An AI agent from Scrapybara. They've got a background in virtual desktops, so this move into agents is one to watch.
Browserbase upgraded their MCP Server - Think of a virtual browser that you can put claude (or any ai) in charge of.
Fellou: Claims to be the "world's first agentic browser."
Granola now lets you chat across all your meetings and it knows me better than chatgpt. My immediate thought: would be even better with project-based folders for meeting chats and granular access controls.
Reply by Nox: AI for iMessage. If I remember right, this helps draft replies or tackles messages you've missed. Could be a timesaver.
Buildship Tools: For crafting and exporting MCP-ready tools for your AI agents.
rehearsal.so: This is a cool find. Lets you practice real-life conversations with an AI, and even gamifies it with “compete with friends” angle.
Socialmotion.ai: A social media video creator. Takes in images from your product website and adds tet with animation over them. No video gen.
Cactus - This tool wants to put your business on autopilot with AI. It has custom sites, voice agent for lead pre-qualification, booking management etc.
Also Betaworks latest AI camp graduates:
Decode - a whiteboarding + code gen tool
Trampoline - RFP management for services companies
TabTabTab - AI everywhere on your computer
Superposition - AI recruiter for startup founders
Hopper - planning tool for scalable AI coding
Afterimage - helping get justice against medical errors and malpractice (founder here)
JigsawML - AI powered cloud management platform (founder here)
NetAssist - making digital world easier for seniors (founder here)
Graze Social - tools to build & monetise custom social feeds
🍦 Afters
Notion has an integration with Replit now. You can now build a website/app with using Notion as your CMS or database.
ICYMI, Y Combinator's latest Request for Startups (RFS) is out.
Zed is another code editor leaning hard into AI features, claiming to be the fastest.
Checking out Robleh’s new AI studio, Tomorrow.
RL with distributed compute - interesting approach from this new “rent a GPU” company.
That’s it for today. Feel free to hit reply and share your thoughts. 👋
Enjoy this newsletter? Please forward to a friend.