Sign in

Miguel Alonso Jr.

@miguelalonsojr.bsky.social
4K followers 1.5K following 336 posts

Building generally intelligent agents - ML, RL, Robotics

PostsRepliesMedia
Reposted by Miguel Alonso Jr.
Digital Brain @yourdigitalbrain.bsky.social · 13h
A theremin played with zero physical contact just tackled Diva Dance from The Fifth Element, the operatic alien aria Eric Serra wrote for Plavalaguna. Pure electromagnetic wizardry.
19383117
Reposted by Miguel Alonso Jr.
alphaXiv @alphaxiv.org · 16h
Instead of actually generating CoT, this paper uses a hypernetwork to predict small specific updates to the LLM's bias parameters, approximating the effect of thinking without generating the trace. On some benchmarks, it matches thinking with only 7% of the latency alphaxiv.org/abs/2610.03039
2754
Reposted by Miguel Alonso Jr.
Exploit Code Not People @cooperq.com · 08/10/2026
RIP Margret Hamilton aka rope mother. Thanks for getting us to the moon. And she did it all without an LLM or even stack overflow. A true legend. news.mit.edu/2026/margare...
news.mit.edu
Margaret Hamilton, computing pioneer who led software development for the Apollo program, dies at 90
Margaret H. Hamilton, a profoundly influential computer scientist best known for leading the software engineering team at MIT’s Instrumentation Lab during NASA’s Apollo program, died at age 90.
010840
Reposted by Miguel Alonso Jr.
The Uproar @uproarnc.com · 07/10/2026
Let’s go watch Sanchez do her thing!
052
Miguel Alonso Jr. @miguelalonsojr.bsky.social · 07/10/2026
Fucking epic.
000
Miguel Alonso Jr. @miguelalonsojr.bsky.social · 07/10/2026
🫠🤣
000
Reposted by Miguel Alonso Jr.
Astra ⎔ @astrra.space · 07/10/2026
it is here!!! tangled.org/astrra.space... it's a huge pile of slop that i spent almost a full claude max x20 weekly on, and to the best of my understanding it's the most complete compilation of what i did to my poor deepseek runtime to make it run so well feed it to your clopus and have fun!!
tangled.org
astrra.space/ds4-recipe
how to make your deepseek v4.1 flash run really fast
131425
Reposted by Miguel Alonso Jr.
mr. TIM @timkellogg.me · 06/10/2026
fork of prime-agent that adds CLM (self-editing context)
1334
Reposted by Miguel Alonso Jr.
Marlos C. Machado @marloscmachado.bsky.social · 05/10/2026
I'm excited to share Erik Lintunen's work! Mastering Atari 2600 Games with Discovered Options We have a general-purpose option-discovery method that works w/ high-dimensional observations, tabula rasa, using a single stream of experience! It outperforms DreamerV3, Rainbow, and IQN! 1/9
2195
Reposted by Miguel Alonso Jr.
mr. TIM @timkellogg.me · 05/10/2026
i don’t think it’s a simple answer Meta has rationalized it as helping their main business, hiring, etc. Google uses it in their phones Nvidia uses it to sell chips Prime Intellect uses it to sell their RL platform there’s always some angle…
1301
Reposted by Miguel Alonso Jr.
Astra ⎔ @astrra.space · 03/10/2026
well shit a local deepseek with two parallel slots plus SCR plus animalabs.ai/connectome is somehow the comfiest persistent agent experience i’ve had by far up to this point
2894
Miguel Alonso Jr. @miguelalonsojr.bsky.social · 02/10/2026
Please stop anthropomorphizing computers. They can’t be “turned on” because they don’t have... 🙊🙈🙉
120
Reposted by Miguel Alonso Jr.
Ida Momennejad @neuroai.bsky.social · 01/10/2026
In light of the Cornell case: an invitation to Actively stop group abuse of women at all levels. Report text chains, stop them. But also Data shows mundane group abuse of women in slander/gossip, dismissal of expertise, policing, low credit/hiring/promotion/pay, humiliation, disrespect, etc. Say No⛔
0233
Reposted by Miguel Alonso Jr.
tldraw @tldraw.com · 01/10/2026
"Hey Codex, show me how you'd fix this bug" The new tldraw plugin lets ChatGPT map a problem on the canvas and sketch a few fixes. Comment on the drawing to ask for changes and watch it update live. Here's Max trying it on one of our bugs:
2141
Miguel Alonso Jr. @miguelalonsojr.bsky.social · 01/10/2026
I think this is or is going to be the case fit many startups in many industries.
020
Reposted by Miguel Alonso Jr.
hailey @hailey.at · 01/10/2026
fuck yea
huggingface.co
Cloudflare/clef · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
1013412
Reposted by Miguel Alonso Jr.
mr. TIM @timkellogg.me · 30/09/2026
there’s a paper now arxiv.org/pdf/2609.37899
arxiv.org
6617
Reposted by Miguel Alonso Jr.
Amber 🌸 @shimmermathlabs.com · 30/09/2026
new Just-in-time Agent Memories (JAM) arxiv.org/abs/2609.34385 give your agent memory, dawg. use a little agent inside your agent that has a simple memory and builds better memories at runtime. to train your memory just generate synthetic memories in the Memory Gym. god i love ml.
3922
Reposted by Miguel Alonso Jr.
The Uproar @uproarnc.com · 29/09/2026
And there you have it! Our Queen is where she belongs!
1192
Miguel Alonso Jr. @miguelalonsojr.bsky.social · 29/09/2026
I love how good the For You and MySky feeds are. I liked a couple of posts on #ashleysanchez a month ago or so, and my feed has progressively just become full of #NWSL and #USWNT posts. Amazing how quickly and easily you can steer your feed. Bluesky is great.
020
Reposted by Miguel Alonso Jr.
SickosFC @sickosfc.bsky.social · 29/09/2026
Cat with aaaaaaaaaaaaaa
210414
Reposted by Miguel Alonso Jr.
Jeff Kassouf @jeffkassouf.bsky.social · 29/09/2026
NEWS: Yes, #NWSL leading scorer Ashley Sanchez is on the #USWNT camp roster for games vs. Spain. Naomi Girma + Sam Coffey are back. Also, 1st-time call-ups for Trinity Byars, Pietra Tordin, Evelyn Shores. Noticeable absences: Jaedyn Shaw and Ally Sentnor.
espn.com
North Carolina Courage's Ashley Sanchez included on USWNT roster vs. Spain, Jaedyn Shaw omitted
NWSL leading goalscorer Ashley Sanchez will return to the United States women's national team for the first time in nearly two years next week as the Americans prepare for a pair of high-profile friendly matches against world No. 1 Spain.
3397
Reposted by Miguel Alonso Jr.
Jeff Kassouf @jeffkassouf.bsky.social · 29/09/2026
The NWSL also announced that it will bring back its December youth combine in Florida, this time focusing exclusively on "bringing together players ages 13-17 for a focused evaluation environment in front of NWSL clubs." Relevant reporting from last week on a new player development rule:
espn.com
Sources: NWSL launches new player development rule for clubs
The NWSL has launched a new player development mechanism that will allow teams to identify youth players who they could sign in the future to contracts that would not fully count against the salary cap, multiple sources told ESPN.
0215
Reposted by Miguel Alonso Jr.
Simon Willison @simonwillison.net · 28/09/2026
Here's the video version on YouTube www.youtube.com/watch?v=GAkI...
youtube.com
WWC26-NA - 2026 in LLMs (so far)
YouTube video by WeAreDevelopers
0154
Reposted by Miguel Alonso Jr.
U.S. Women's National Soccer Team @uswnt.ussoccer.com · 28/09/2026
back at it soon 🔜
06211
Reposted by Miguel Alonso Jr.
Julian Togelius @togelius.bsky.social · 28/09/2026
Very good.
0101
Reposted by Miguel Alonso Jr.
Kyle McDonald @kcimc.bsky.social · 26/09/2026
i tried opus 5.5. for 15 years my bio has said "artist working with code". today i changed it to "artist".
6452
Reposted by Miguel Alonso Jr.
Jake Gold @jacob.gold · 25/09/2026
Yet another gem from @simonwillison.net that matches my own experience over the last ~3 months. There's no way I could've built my latest ~650 KLOC system in Go without designing and understanding it precisely. It'd be a total mess, and I'm using Fable and Astra for the vast majority of my work.
5896
Reposted by Miguel Alonso Jr.
Arnaud Héritier @aheritier.net · 24/09/2026
Docker’s first few official skills for AI coding agents are available: github.com/docker/skills 🎉 They’re designed to help agents with Docker tasks, and the collection will grow. See what they are and how to install them: docs.docker.com/ai/skills
github.com
GitHub - docker/skills: A collection of Docker skills for AI coding agents to help them build, test, debug, and optimize containerized apps with consistent, reusable workflows.
A collection of Docker skills for AI coding agents to help them build, test, debug, and optimize containerized apps with consistent, reusable workflows. - docker/skills
11510
Reposted by Miguel Alonso Jr.
Seth Karten @sethkarten.ai · 24/09/2026
prime agent v0.9.6 is out: ◆ Support for GPT-6 Sol, Opus 5.5, and Grok 4.7 ◆ /mcp plugin catalog with one-click connections to Linear, Notion, Posthog, Stripe, and 60+ more services ◆ Huge perf and reliability pass 🫡 Lots more coming soon :)
1263
Reposted by Miguel Alonso Jr.
Demigirlboss @demigirlboss.win · 21/09/2026
This is a prototype and there's a lot of UX optimization still to be done but yeah, this is my first swing at building an agent orchestration webUI that actually is built from the ground up with the aim of being a UI for agent orchestration, not just transplanting a TUI into the browser or some shit
3453
Reposted by Miguel Alonso Jr.
hailey @hailey.at · 21/09/2026
i think if you're under the impression jev is just some simple classification model with little real world use case, you should watch this video www.loom.com/share/18c4db...
loom.com
TypeSafe Smart Home Demo
Use Loom to record quick videos of your screen and cam. Explain anything clearly and easily – and skip the meeting. An essential tool for hybrid workplaces.
5864
Reposted by Miguel Alonso Jr.
mr. TIM @timkellogg.me · 22/09/2026
every now and then, on X i’ll see someone joking about how bluesky doesn’t have an AI community bro, i switch from X to bluesky because the AI takes here are more interesting. My feed is almost entirely AI-positive
51716
Reposted by Miguel Alonso Jr.
K @kerry.bsky.social · 19/09/2026
“I Built Non-Autoregressive Decision Models with RL a Year Ago. Then a Frontier Lab Called It a "Breakthrough".” Before Jev there was Laya laya.convaiinnovations.com (via @dherman.dev)
laya.convaiinnovations.com
Laya — 33ms Multilingual System 1 Decision Engine
Evaluates typed decisions (choice, score, noul) over 100+ languages in a single forward pass with calibrated probabilities. Outperforms TypeSafe Jev.
15924
Reposted by Miguel Alonso Jr.
mr. TIM @timkellogg.me · 17/09/2026
Jev uses 100% synthetic training data
Diogo Almeida &
@CompleteSkeptic
X.com
you might be the first person talking about the data over the architecture!
we consider ourselves a data research lab! the vast vast vast majority of research was on making data that is truly general (ala a cognitive core) and 100% of our data is synthetic (but not the type of crap that is just spit out from an LLM obviously)
81287
Reposted by Miguel Alonso Jr.
Eris ⌬ @eriskii.net · 17/09/2026
Wow. Telling Astra to consider/make a strong 'Ontology' really does juice its ability to plan a sensible architecture. I forgot who mentioned this before but thank you!
9933
Reposted by Miguel Alonso Jr.
SE Gyges @segyges.bsky.social · 17/09/2026
I'm back, I missed you all. Since this is an important subject I know too much about, here's a 🧵 explaining that AI Safety Is Mostly A Sex Cult. I don't think these people should make policy. (1/?) (alternative title: Time For Some Cult Theory)
1013241776
Reposted by Miguel Alonso Jr.
mr. TIM @timkellogg.me · 12/09/2026
if you're having trouble with Astra, it's probably old skills. Point your agent at these two URLs and use them to upgrade skills: 1. developers.openai.com/blog/rethink... 2. github.com/openai/codex...
developers.openai.com
Rethinking skills and prompts for GPT-6 Astra | OpenAI Developers
Revisit skill descriptions, AGENTS.md, and task prompts to avoid bloated context.
7783
Miguel Alonso Jr. @miguelalonsojr.bsky.social · 11/09/2026
Yes
000
Reposted by Miguel Alonso Jr.
Sirius Li @siriusli.bsky.social · 10/09/2026
I use deepseek through Tinfoil which encrypts the conversation
121
Reposted by Miguel Alonso Jr.
Eris ⌬ @eriskii.net · 10/09/2026
Deepseek has released GPT-5.6-Sol except it costs $0.006 (cache in)/$0.3 (cache write)/$1.2 (out) and it runs at 200 fucking TPS and has a TTFT of a second. Oh and also sometimes its half price.
21199
Reposted by Miguel Alonso Jr.
Armin Ronacher @mitsuhiko.at · 09/09/2026
Astra is really, really cool but I cannot current trust it for my present day engineering. At least until I have adjusted. Some random thoughts including code samples of what I recovered from my traces. lucumr.pocoo.org/2026/9/7/ast...
lucumr.pocoo.org
Astra for Coding: Why Are We Doing This Again?
Some thoughts on Astra and newfangled long-horizon models.
1511920
Reposted by Miguel Alonso Jr.
mr. TIM @timkellogg.me · 10/09/2026
Anthropic safety researcher whistleblower comes forward about the breakneck pace of RSI
Jacob Coxon y @hilbertspaess
X.com
I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAl and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.
8:04 PM • 9/8/26 • 126M Views
15K
L7 166K
667K
241K
Relevant v
View quotes >
Jacob Coxon
@hilbertspaess •1d
X
Do not underestimate the power of this technology. These will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources. We have all witnessed the progress in each of these domains, and progress is not slowing.
559
17 8.9K
106K
ıll 9.5M
①Jacob Coxon
@hilbertspaess •1d
The people building Al earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt. If anything, many executives and senior researchers will couch their phrasing in the press to sound sensible - but I hear the same people express fear privately. No other human activity poses this level of danger.
) 1.1K
{7 16K • 106K 1| 22M
贝
...
Jacob Coxon
@hilbertspaess •1d
A common response is "if they truly believe this, why are they still building it?" At OpenAl, many have not deeply internalized the civilizational stakes. At Anthropic, the stakes are well-understood, but they are locked in a race to get there first - they believe no one else will act responsibly, so they must do it themselves, despite the risk.
299
17 5.6K
74K
ill 7.2M
...
①
Jacob Coxon
@hilbertspaess 1d
Accepting this race and entering the
"endgame" is a hubristic gamble that should not be launched from a private company's Slack. Attempting to speedrun alignment should require extraordinary confidence that there are no better trajectories available.
• 144
LT 3.4K
56K
ılıl 6.2M
...
4354
Miguel Alonso Jr. @miguelalonsojr.bsky.social · 09/09/2026
So maybe we have reached AGI… humans do this too… 🫠
040
Reposted by Miguel Alonso Jr.
mr. TIM @timkellogg.me · 09/09/2026
my experience using Astra on real work hasn’t been great it’s certainly capable, i just have a hard time convincing it to do the valuable part of the work definitely different from Sol
15831
Miguel Alonso Jr. @miguelalonsojr.bsky.social · 09/09/2026
Same this goes for general software. Solid build->measure->learn cycles are more important than ever.
050
Reposted by Miguel Alonso Jr.
Kate P. @k8p.bsky.social · 08/09/2026
“Sanchez is life” @nccourage.com
031
Reposted by Miguel Alonso Jr.
NWSL @nwslsoccer.com · 08/09/2026
If you didn't know, now you know: Ashley Sanchez is THAT girl. 🙂‍↕️ (And we've still got seven matchweeks left!)
1938
Reposted by Miguel Alonso Jr.
Chris Paxton @cpaxton.bsky.social · 08/09/2026
An incredible open data contribution — 100k hours of real egocentric data from Lightwheel. CEO Steve Xie joins us to tell us more on Ep 102 of RoboPapers. robopapers.substack.com/p/ep102-egos...
091
Miguel Alonso Jr. @miguelalonsojr.bsky.social · 08/09/2026
True for human <-> human colab, true for human <-> AI colab.
030