-3D Dev+AI Sommelier “Be useful to society.”-@ScottAdams “Analyze Competitors and Win.”-Astra
Pennsylvania, USA
Joined January 2026
- Tweets2.2K
- Following92
- Followers167
- Likes1.4K
Pinned Tweet
My first conversation EVER with GPT started with a question:
“What’s scarier?,
1. We travel to the end of space and find an impenetrable wall (The end of the universe)
Or
2. We travel to the end of space and find it’s infinitely large (infinite number of universes)”
I haven’t put GPT down since that conversation…
This couldn’t be more true,
I felt guilty when I signed up for the $20/month subscription in April.
Now I spend $200/month and am considering getting more.
(This is coming from someone who’s worn the same pair of Jordan’s every day for the past two years)
On average, my current usage of Codex amounts to between 700 million and 1.8 billion tokens per day.
The $200 subscription plan is no longer sufficient; I’ll likely have to upgrade to the $500 tier.
I never could have imagined just how much AI would truly change everything, permeating every aspect of life and transforming the way I work.
And yet, we live in an ivory tower. 95% of people use the free tier of ChatGPT, while 5% pay. And of that 5%, only a small fraction are "hardcore" users like us.
Apparently Astra loves implicit state machines,
Try it out, Astra immediately found multiple per app, even in my new smaller ones.
The thing about Local models is, Qwen 27b is as good as it gets,
Yes, I know Deepseek does better on benchmarks.
The issue is, the difference in actual quality of work is negligible for the difference in parameters.
(As in you can’t ship either result as is, sure Deepseek or Kimi can do distinctly better.)
But you want to do real Art/Game rendering? You’re going to need a Fat GPT or Opus+.
Don’t go into local models thinking you’re going to “run GPT for free” unless you’ve got solar panels, a good way to keep it cool and quiet.
(And an extra 30k to run a potato by Fable standards)
Be patient, wait for a real AI workstation with a legit chip in it like @OpenAI jalapeño.
Use the frontier or a small Qwen for now, trust me, no need to get 30k in substitute hardware (GPUs)
Imagine being someone outside of X and looking at this chart.
They’d likely say something like:
“20% higher big deal”
AI goes into Humanoid (logically)
Humanoid goes into Workforce (logically)
Human goes into…(Extinction??)
Take robo taxis into consideration, Humanoids accelerating at the rate AI accelerated will be a wild time.
Don’t get me wrong, we’ve got a longgg time until robots are ready but still…
Sooo you can have 3 humanoids for the price of 1 human worker…
Tell me how this is going to work in 10 years…
The universal income thing is going to fun I guess😅
Asimov 1 is an open-source humanoid robot you can build, train, and customize.
- 1.2 m tall, 35 kg
- 25 powered joints
- Open hardware with off-the-shelf parts
- Open-source SDK and simulation models
Available as a DIY kit.
Order yours at menlo.ai/order
This is interesting!!
Tbh I never saw any “state values” out of Qwen when I was running it.
And I’m curious how the results differ to a more surgical ‘abliteration’ approach,
But this makes me more optimistic for Open weights customization!!
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
Heads up all Codex users @ajambrosino has been teasing he’s the new reset Czar,
Make sure you have his notifications on!!
New prompt for you, use during polishing:
“Does everything look scalable? Do we need to use Bend for anything critical to make sure it doesn’t break for 1000+ users”
⏰ Super excited for this!
There’s going to be some super talented people here working at an Ultra Accelerated pace!!
Get ready to Lock-in.
With the right prompt, $50K could be yours: epic.gm/epic-ai-megajam-2026
#MegaJam
Hands down my largest post ever with almost 1,000 likes.
It took me 4 seconds to write. I almost didn’t post it because quote tweets don’t do well on my account.
Maybe because i mentioned Open source I got the attention of a new set of people🤷♂️
If my actual work could get this much traction I’d be very grateful🤣
Not a great sign, but in the grand scheme of it,
-OpenAI just went on a $30 Billion funding campaign (secured ~11 billion).
-Anthropic burning $80 Billion in 1/2 year.
$500 million is a sellable figure at this scale.
BREAKING: A mystery seller is trying to dump about $500 million in data center loans all at once, per Bloomberg, days after the midterms raised doubts about many projects.
The anonymous seller is auctioning a dozen loans, including a $25 million piece of financing for a huge Stargate site in New Mexico, the joint project of OpenAI, SoftBank and Oracle.
Oracle has already sent that site’s developer a force majeure notice, a legal claim that events beyond its control may excuse it from its obligations.