The Nevada Department of Transportation has a live feed of traffic cams, so I quickly Codex’d a page which renders all the cams on the Las Vegas GP F1 track
The UX could be improved, but I'm late for my run. GitHub link in thread, just download and run index.html locally
It turns out it's because we were literally asking GPT-5 to output the date after the item! It was just following our instructions
3/4 🧵
1
26
Great take from @btaylor on @AcquiredFM: Say you are debugging code that caused a system to shutdown- you don’t just restart the system. You find the part of the process that was broken and caused the shutdown.
The same philosophy should apply Cursor/Codex/Claude. If it produces incorrect code, the philosophy should be don’t fix the code, fix the context that Cursor had that produced the bad code
If you just fix the code you don’t have leverage. If you go back and say, ‘what context did this coding AI not have that if it had it, it would have produced the correct code?’
It takes longer in the short term, but in the long run it’s the difference between properly leaning into AI vs just using AI
1
256
Sharing my GPT-5 tl;dr that I sent internally after watching the livestream at 3am Sydney time:
Cursor/Coding
- GPT-5 is out in Cursor today and is their default model
- Early testers say it’s very good at pair programming, better than Opus 4.1
- Lots of people reporting it’s very good at very long conversations/pair programming sessions. But say Opus 4/4.1 still wins on agentic coding tasks
- It looks fast - much faster than o3, and gpt-4 when it first came out
API
- There are 4 models: gpt-5, gpt-5-mini, and gpt-5-nano, gpt-5-chat. I’m particularly to see what gpt-5 is like vs gpt-5-chat. When we moved from GPT-3 to GPT-3.5, it got a lot better at chat tasks but lost some abilities in the RLHF process
- It’s available in the API today
- It decides how much to think/reason on a problem. You set this in the API with the reasoning parameter, which has a new minimal reasoning param
- There is a new verbosity parameter (low, medium, or high)
- Structured outputs can now check regex (previously just a JSON schema)
API Pricing:
- Input tokens are 50% cheaper than gpt-4o
- Output tokens are the same price as gpt-4o (but with thinking, it would use more tokens)
ChatGPT App
- GPT-5 decides how hard to think on a problem by default. This unlocks reasoning for so many people who didn’t use or have the model picker
- OAI are deprecating all the other models
- Early testers say it’s better than o3 for similar research (where simple = you don’t need to go deep into a problem, not that the problem is hard) but not as good as o3 on deeper research. (I hope that’s not true, since I use o3 to dive deep ~20 times a day)
- They say a big unlock is that it is faster. Letting you iterate much faster at a high quality each cycle
1
285
GPT defaults to returning JSON pretty-printed, with indents & new lines - great for humans but redundant for computers.
We found asking it to output minified JSON reduced tokens significantly, cutting total response time by ~30%
This saved us 30s on a call that was 70s (TTLT)
1
125
The best PM and best dev are the same person…
It’s becoming increasingly necessary to hire engineers who have strong product and design sense, and PMs who have at least foundational technical knowledge, e.g. programming 101 at uni
133
2016: computers can talk (WaveNet)
21: computers can code & recognise images (Codex, GitHub Copilot, CLIP)
22: computers can chat & recognise speech (GPT-3.5, Whisper)
23: computers can see (GPT-4 Multimodal)
24: computers can reason (o-series)
25: computers can ?? (operate/act?)
1
109
A great way to cap off my first WWDC. As someone in AI/ML, it’s been fun pretending to be an iOS developer for the week! 🥸
Lovely spontaneous #ios dev hangout at Super Duper Burger, #SanFrancisco. Probably the last one for #WWDC this year as ppl are starting to head back home…
Thank you everyone who joined for dinner, tech talk and silly card games generated by #chatGPT 🥳
1
4
419
Really enjoyed my first live @thetalkshow!
A lot of Mac, Apple Silicon, and Vision Pro discussion with @gregjoz, John Ternus, and Mike Rockwell, but I found @craigfederighi's answers about Apple's approach to AI/ML particularly interesting, especially since they have been a bit lost due to the Vison Pro.
Some noteworthy notes and quotes...
AI/ML
• ML supports so much of the experience across their devices, but “it's the experience that is the headline, and the technology is all in support of that”
• On UX: “[Using ML to take a picture of a PDF, figure out where all the fields are, suggesting what to type into them” is like magic paper! It's automatic, and you aren't having to think about it, you're not chatting with the thing. It's automatic.”
“[When you have a PDF form, you see the fields and know how to use them and what to enter.] Now the device can see the form, and it knows what to do for you. That’s how you should view [AI/ML]. They're a transparent part of meeting the most natural expression of your experience, of your intent, when you're using the device.”
• On staying quiet: “If you look at any area of technology, I think what we've seen is whether it was the microprocessor, bitmap displays, touch technology, the internet, all of these things, in the end, had a lot of experimentation going on in public. And then, in the end, were great areas for Apple to figure out the right way to deliver compelling customer experience.”
• On privacy: “No, Apple's state on privacy is not going to slow us down… the data that needed to train these models is is publicly available, not personal data. We do not need your personal data to make our system smart, and when we need to get specific data we're not doing that by spying on people. We're going to go out and get it the right way, respecting users' privacy”. In other words, Apple aims to develop great, generalised models.
• Alluded to was that Apple Silicon will be a key differentiator in allowing personal data to be incorporated, both as context and in fine-tuning
AR/VR
• They are really positioning this as a standalone device, not just an extension of the computer or phone
• Strong emphasis on making sure the software experience feels right at place in the environment, rather than taking you away from the world
• No mention of hardware specs, other than the display was akin to a retina display, meaning the pixels imperceptible to the naked eye
Other
• The Apple ecosystem will only get better as barriers continue to break down. Craig: “we want each platform to be the best embodiment of what they can be, but in a way that is as consistent as possible. Every year the movement from one of my devices to another is increasingly without thought, your data's there, the experience is consistent, even the way that different gestures and actions work… We want the entire experience to feel like it is the work of a single designing hand that created the right Mac experience and the right iPad experience, the right Apple Watch experience, and so forth, but with a common set of values across all of them.”
• Craig’s still got it ✔️
From their Q1 results: “We are pleased with the current performance and trajectory of our per-member advertising economics. In the US for instance, our ads plan already has a total ARM (subscription + ads) greater than our standard plan.”
1
57
William Ritossa retweeted
Today @OpenAI released an IOS app. 99% of you are using it wrong. Follow this thread to learn more...
Adaptive governance: given this is a field is rapidly evolving, this is obviously apt. Covid seems like a close model for this however in most parts of the world states had to enact emergency powers to really be adaptive. The last thing anyone wants is layers of subcommittees
1
66
Incentives for compliance: this is intriguing! Does Sam mean minimum safety/reporting requirements, alongside 'stretch goals'?
Perhaps this could look like tax credits for safety research, independent audits, etc. Or something more novel (compute, quicker renewal process). It could become like SOC 2 where it isn't *technically* mandatory, but in effect essential for work with large clients/orgs.
Alternatively maybe he means a more proactive and positive approach, rather than punishing non-compliance. That seems much more radical and like it would lack political capital.
60









