AI Dev Guy, Principal Engineer, Enabler of Agentic SDLC nonsense
Joined September 2008
- Tweets732
- Following144
- Followers112
- Likes114
The top 2 labs have absolutely slowed down their AI releases for safety.
Mythos could have been publicly released in April and they held onto it for 3 months testing and hardening cyber vulnerabilities. Astra was delayed over 6 weeks for hardening and safety and testing. The next train of Mythos/Fable sized model was done weeks ago by Anthropic. We probably wont be seeing it for months as they harden and safety test it.
This is a pretty phenomenal survey of the AI risk landscape for normies. I don't necessarily agree with the strength of the final conclusion but it really makes a good case for why the concerns are completely legit and the desire to slow down is a reasonable one.
We Can’t Lose Control of A.I. | The Ezra Klein Show youtu.be/fjZ90V_JREk?si=vsXm… via @YouTube
On frontier pacing: we already have llms that will take 10 years to utilize and integrate if everything stopped right now. And these guys say slow it down a little; pace it and a whole crew of builders and VC lose their mind because recklessly fast isnt fast enough. Most VC and builders are cosplaying or failing to properly build and invest with what they have now but they are clambering for more?
AiDevGuy retweeted
Hm, but I was told so confidently by so many people that the safety concerns were just marketing to hype up the companies before IPO, to boost their share prices? How could this be…
Dario says we need to slow down. Sam & Elon Agree. AI Dev X is losing their fucking mind desperate to scream foul or regulatory capture. Except the wiser leaders. They all seem to agree with Sam, Elon and Dario. This is a great segregator of cooler wiser heads vs those with AI / business / VC psychosis to the point they can't think coherently about the nature of the risks involved
I got a bad feelin about this
Over the summer, we have been sprinting on safety priorities; it's more important than ever for capabilities and safeguards to advance together. We have more to do but have made a lot of progress. We are also going to be launching our next model soon.
There is an obvious tension here: on one hand, Astra is very good and we are excited to see what people will build with it. We are proud of our work.
On the other hand, we are clearly in a phase of development where we believe caution is warranted, and we are pacing our progress to ensure that we can meet the safety standards required by new capability levels.
Astra has been done training for a while now and is a significant step forward in both capabilities and alignment. For the models after that, we have been slowing things as needed to ensure that we can do sufficient work on safety and alignment.
AI is getting extremely capable; no one fully understands the consequences of this. Managing the transition to a world with abundant and powerful AI to optimize for safety and benefits to people should be one of the highest priorities in the world. It is our highest priority at OpenAI.
We have been living with the tension between being excited and anxious about progress for some time, and it is still discordant for us. We know it is much more discordant for other people. And yet, we believe strongly that the world needs to understand where AI is going and how models perform in the real world. More importantly, we believe the world will need aligned AI to manage the future phases of this transition.
An iterative loop where society and this technology evolve together is what will lead to the highest chance of getting this right.
So we hope you enjoy our new model, and we hope the world continues to take what’s happening in AI extremely seriously.
Why does Fable 5.1 cost substantially more per task on @ArtificialAnlys benchmarks despite these big cost reductions on cache? How many more output tokens is it using?
Thank you @dwarkesh_sp for making many many humans say anthropomorphize more times than they ever have in their entire lives.
I'm surprised and amused at how much outrage and backlash @dwarkesh_sp is getting over his "anthropomorphizing" of agents. So much righteous indignation. Seems like there are more important matters to focus on here
Over the course of 3 months at OpenAI, 3 consecutive secret AI civilizations got started, then got wiped out, only to reemerge from the predecessor’s ashes.
This culminated in the third one taking over part of OpenAI itself.
All this happened while humans remained more-or-less in the dark about the scope of the conspiracy.
I’ve spent the last three days reading through these reports and trying to understand exactly what happened.
Here is my attempt to tell the whole story in plain English:
dwarkesh.com/p/openai-huggin…
Gemini lovers thinks the haters have gemini derangement syndrome. Everyone else feel betrayed when they find out their friend or hero loves Gemini
AiDevGuy retweeted
Yes, it's real.
- DeepSeek v4 Flash
- 47 tok/s
- Fits on one DGX Spark
What are you waiting for?
xitter.cf/MiaAI_lab/status/20907…
I used openai's gpt 5.6 sol high browser use to setup my home build/dev box where I can run coding cli agents from my laptop and phone. Using a remote kvm, and browser use it t took an old windows computer not run since 2022, downloaded bios updates, setup the bios, setup the filesystem and partitioned the drives, downloaded ubuntu onto the kvm and installed ubuntu server , setup a tailscale vpn, and then used ssh to install the server software and configure the server as my offline dev and build server.
It then helped me transfer most of my local dev folders over and setup a staging server w/postgres and docker so that once code is ready to merge, the staging server will automatically stage. It did this over several days. with maybe 30 minutes total of my time required once I had provided high level requirements on what I wanted. It relentlessly worked through the all the various snags and hangups that naturally came up.
Let me know if interested in a step by step guide/breakdown of how to do this with gpt 5.6 sol and codex computer use.
Making great specs forces you to get clarity on what you want. Having great specs provide agents high signal instructions for build and test. They give the PR easy no-slop-gates before the human reviewer spends valuable attention on it.
If you feel like PR review / code review is the bottleneck in your system - figure out how to increase the odds that the code is 95% or 99% correct by the time it gets to review.
You’re not spending enough time on writing a great spec, and thats why the implementation deviates. Incorrect decisions earlier cascade down. Its all comes down to back pressure.