Anthony

I keep encountering this weird bug from OpenAI Realtime models where it either stops the lessons early or just freezes after a while. I thought it went away when I switched it to 1.5 but it's happening with 2 as well.

@MsEscoTeaches Too relateable

Here’s the sad thing about reading… I teach a phonics intervention class for 30 mins a day. It is extremely difficult to get them to pay attention, practice, and participate. They fight me every step of the way. I’m at a loss on how to help them when they won’t help themselves.

@sistilli You were Canadian this whole time?!

we have some of the best engineering talent leave canada every year because building in Canada is horrible fix the leak before adding more water

@TheBaamAhmed I'll post one soon! (ish)

@antho_builds yes! i think i've seen you on my tl a couple of times as well, would love to get on a call and trade demos

if you're a student at waterloo doing a summer academic semester, i need your help i'm exploring this idea of an "AI TA" (rough demo below), and to know if the concept works, i need to test it in a real academic setting so here's my proposal: you get: unlimited access to @ask_theta for the duration of the testing + i'll buy you a coffee + donut i get: 20 mins of your time twice a week for some feedback

@sutherlandphys Just curious... are you still hiring?

2 (waterloo) interns started here yesterday at PhysicsGraph, got onboarded and started contributing legitimately useful stuff right away very promising

@noahkostesku How many students lose confidence in their math ability due to bad teaching? That's why I'm building an AI Math Tutor.

Looking for people in Toronto who are unreasonably obsessed with AI. Comment what you’re working on or dm me

Thanks to Harj: youtu.be/57lDpTwiW6g?si… I now realize I'm not operating from an abundance mindset and I'm too afraid of using my money.

I do get sad knowing I'm going to miss a lot of significant events happening in Toronto while I'm in my youth. 2024 - Candidates Tournament 2026 - World Cup I just feel like I haven't earned the right. But then I think about it for a second and it's barely a sacrifice lol

@kwindla @dan_jenkins I'm building an AI Math Tutor that needs to be able to add equations to a whiteboard (tool call) and then explain and asks the student questions and interprets their response. Should I be using a S2S model or building a pipeline?

It's not even clear the you lose nuance. Speech-to-speech models are better at some kinds of audio understanding and worse at others, today. The obvious, easy choice is if you have a mixed-language use case that has short sessions and doesn't need much tool calling. Speech-to-speech has much better audio understanding and more natural speech output, for that case. For almost everything else, even if you're focusing on nuanced audio understanding I think you get pretty mixed results from both vibe testing and formal benchmarks.

S2s models are the end goal. But for multiple reasons, as a consultant, I wouldn’t move a client onto one now. Building a pipeline offers far more flexibility for very little added latency; sure you lose a little nuance… but the end result will be better right now

I've decided that I should spend more time with gpt-realtime-2 for my AI Math Tutor. The old model needed a lot of scaffolding and there's a clunky feel to the conversation.

@ben_m_somers Applied

We are hiring teachers for $100,000-$130,000 per year based on experience for an exciting research study Please check out modernbloom.org to apply!

Lack of clarity is absolutely killing my productivity.

@mark_k @OpenAI Will this be available via the Realtime API anytime soon?

A new “voice mode” is being prepared for release by @OpenAI. The upgraded voice mode is based on the omnimodal GPT-5.5, making it substantially smarter and more expressive than the current version. It will also support full-duplex conversations, meaning it can listen and speak at the same time. That should make conversations feel much more natural and fluid.

@p_kuperman Hey Peter, I'm an educator from Brampton who has been fascinated by Alpha School this past year. Are you looking for more educators to help launch Alpha School in Toronto?

I am helping to launch an Alpha School in Toronto. Reach out if interested in learning more!

I'm seriously lacking clarity right now. Torn between working on the AI Math Tutor and tutoring vs switching directions.

@p_kuperman This is amazing news!

@LifespanDevelop I do agree with a lot of what you wrote, especially: 1) current education system is wasting people's potentials 2) motivation is a key component that is rarely discussed What are your thoughts on Alpha School? Is their system not end-to-end enough?

Honestly, thank you for asking. This is a brain dump so its not very well organized. But here are a few: Education is broken. Here are some key reasons: 1. The fundamental issue with Education is that is not aimed to the right purpose. Education has one key purpose: To develop incredible humans. Everything else, is built either on top of that foundation or a distraction. 2. What we teach humans to become their incredible selves is wrong. The topics, the content, the methods, none of it is organized into a coherent framework that is designed to develop incredible humans across their lifespans. 3. Humans have one key ability that no other being, that we know (maybe there are aliens who do this lol), of has: we take information from the abstract/ether and can rearrange atoms to build those ideas into physical changes to our reality. These nodes of knowledge, these topics, are how we contribute to the universe. These nodes are also the topics we use to teach our students. For example. we discovered Algebra we teach kids algebra. But currently we are teaching incredibly stale topics to our students. At best the information students are learning are years old. We need an end-to-end system for refining knowledge into truth and teaching that knowledge the moment it is discovered. In real-time. The more truthful the knowledge we use to teach humans, the more we can develop incredible humans. The more incredible humans, the more we refine knowledge into truth. This is, in my opinion, our purpose. The Answer: I spent 10 years building three things. 1. A blueprint that reorganizes how we do science and education to refine knowledge into truth and develop incredible humans across their lifespans. 2. A tool: Think a learning management system like blackboard or canvas, and science management system, that enacts the blueprint, so its incredibly efficient for teachers, students, scientists etc. to develop incredible humans and refine knowledge. 3. A campus: an environment designed around developing incredible humans and refining knowledge. There are a lot of people doing incredible things in education and science that can be plugged into this blueprint. Then there are a lot of people totally wasting their time. It drives me insane. And its my fault if I cannot convey how and why they are wasting their time and what they should be doing instead. For example, people are freaking out that students are using ai to cheat, so their approach is to stop students from cheating by making tools that do this: such as ai cheater detectors, browser lockdowns etc. Now these cheater detectors are still useful, because there will always be a subset of students who are motivated to take shortcuts, but what we are experiencing is WAY beyond just a few students trying to gain an edge. but these solutions don't address the first principle problem: Why are students cheating. BECAUSE EDUCATION IS A WASTE OF TIME, IF ITS NOT DEVELOPING THE STUDENT INTO AN INCREDIBLE HUMAN (I'm not yelling at anyone in particular, just into the void lol). And guess what, kids know this. So they should cheat, because education is cheating them. Other problems and solutions: 1. all tasks and productivity apps suck. They suck because none of them implement motivational science. I have a mockup of the entire UI for exactly how to create a task app that is personalized to anyone, easy to use, and collaborate with, and motivates you to strategically accomplish your goals. 2. Grading sucks. We developed a mockup of the UI for a system that applies learning, measurement, and motivational science to provide high quality real time feedback, and grade anything, especially students' high creative products such as essays or projects. The system will have an ai that follows principles to suggest a grade for the paper with feedback comments and the teacher just needs to approve or edit the comments. After about 5-10 papers the system should be accurate and reliable so the teacher is basically clicking "approve" all the way down. So if you made it to the end and want to talk about any of these. Send me a dm. This is the best I can currently do via text.

I feel crazy to say this but my current frustration is that I have too many answers to too many problems and not enough time to explain these answers well enough to convince people with money to fund me to use the answers to solve all these problems. This post will age well.

I was so sleep deprived yesterday that yesterday I made an error that I only figured out this morning.

I just realized that I love what I built. My AI Math Tutor without proper prompting is shit. With proper prompting is amazing.