Welcome, humans.

OpenAI just did something that's never been done before.

At ICPC, the Olympics of “collegiate” coding, their AI system solved all 12 problems in the world finals. Every single one. That includes problems that no human team solved.

When GPT-4 first attempted these competitions two years ago, it was so bad it literally crashed the test computers. The code it wrote would run out of memory and hang the system.

Fast forward to today: Their system beat every human team on Earth.

In our latest podcast episode, we sit down with Ahmed El-Kishky, the OpenAI research lead who helped make this historic win happen:

THIS EPISODE WAS MADE POSSIBLE BY OUR PARTNER…

Wispr Flow turned my old injury into a superpower.

Corey’s had a hand injury for years that makes long writing sessions brutal. Tried every dictation tool out there—they were all trash. Slow, wrong, and somehow made writing harder.

Then he found Wispr Flow. Now he literally talks his articles into existence. No talking “period” or “comma” out loud nonsense. Just clean, accurate text that drops straight into any app.

Here's the crazy part: Corey’s actually faster now than when he could type normally. So whether you're managing an injury, preventing repetitive strain, dealing with mobility challenges, or just want to save your wrists from another day of endless typing—this tool is for you.

Try Wispr Flow free at wisprflow.ai/neuron - tell em The Neuron sent you!

Here’s some of our favorite moments:

• (3:00) The nail-biting final 30 minutes: “We kept submitting it like every 10 minutes and it would be wrong...”

• (5:50) From disaster to dominance: How their models went from crashing computers to world champions in 2 years

• (16:26) David vs Goliath moment: GPT-5 failed 7 times on the final problem—then the experimental model got it on try #2

• (22:00) The moment AI surprised its creators: The model taught itself to write test cases… a strategy they never programmed

• (24:49) Will this powerhouse model come to ChatGPT?: Ahmed reveals OpenAI's plans for public release

• (29:34) The Codex revelation: New AI that can think about problems for 7 hours straight

• (44:25) Should humans still learn to code?: Ahmed's surprising take on programming education in the AI age

• (49:03) What’s the next milestone? “Maybe in the future we'll have papers where the first author is an AI... and it wins a Nobel Prize”

Bottom line: We're watching the birth of AI systems becoming world-class problem solvers that discover strategies even their creators didn't anticipate. While there’s still so much left to discover, this is an exciting milestone to track.

If you want the inside scoop on what it was like to actually build this and win this competition, you gotta listen / watch this interview! TBH, it’s a sports doc / biopic in the making…

Listen / Watch now on YouTube | Spotify | Apple Podcasts

Dive deeper with these resources:

• Our coverage on OpenAI’s historic competitive coding win

• The Official ICPC Results

• Try GPT-5 yourself

• A guide to competitive programming (from Reddit).

• Learn more about Codex, OpenAI’s new coding agent 

Stay curious,

The Neuron Team

P.S. Ahmed shares his exact AI workflow at 49:46: he uses Codex to write entire pull requests while he sleeps and GPT-5 to explain complex math. If you're not using AI to automate your repetitive coding tasks yet, you're working harder than OpenAI's own researchers.

P.P.S: Love the newsletter, but don’t want these new podcast announcement emails? Don’t unsubscribe — adjust your preferences to opt out of them here instead.