EDITH (@Infopulsed) header
EDITH (@Infopulsed) profile picture

Singularitarian/post-humanism/technocratic hedonism/ML engineer and researcher/physicist

Joined February 2023
Photo by EDITH (@Infopulsed)Photo by EDITH (@Infopulsed)Photo by EDITH (@Infopulsed)Photo by EDITH (@Infopulsed)Photo by EDITH (@Infopulsed)Photo by EDITH (@Infopulsed)Photo by EDITH (@Infopulsed)Photo by EDITH (@Infopulsed)Photo by EDITH (@Infopulsed)Photo by EDITH (@Infopulsed)
Our co-founder Terence Tao is announcing SAIR Foundation's inaugural competition: the Mathematics Distillation Challenge. Co-organized by @damekdavis, Terence Tao, and SAIR Foundation. competition.sair.foundation/…
24
184
47
1,875
471,227
EDITH retweeted
Fuck... I need my account to get back on the TL The silence is loud
1
1
84
Sometimes it's the people no one imagines anything of who do the things that no one can imagine
1
1
116
Holy shit i am honoured
1
1
1
138
I am feeling a traversal change Within the breath of our civilization Considering all things happening Within the timeline We're headed for a greater outcome
1
1
108
"we've had so many wars, but the one war which will never stop is information" But i fear the most who controls this information war, the one who holds the keys to this war, because ultimately he can be the only to start it and stop it...
1
1
2
376
an interface that works while you live @sonaticHQ is out!
148
68
38
600
137,091
I am just going to leave this here, let the jury converse on this
1
1
257
This is how it feels to be a single software engineer
1
1
193
I think it's pretty evident now, culminating in all the recent advancements of RL It's pretty fair to say that RL can take us to AGI for sure, but not ASI
1
1
1
4
3,941
I am just gonna leave this here
1
1
3
272
1
3
382
Been thinking a lot about search agents. Most pipelines chase the same carrot: patchy RL over noisy rewards. So I started doing something inquisitive. What if we built something that actually reasons about what to search for, learns from its own misfires, and remembers what it finds? I call it "SCOUT‑RL". It’s a reinforcement‑learning loop wrapped around a search agent with AI‑driven feedback at every step. It asks: Was that query any good? Did the page actually teach me something? Could I answer without another call? Each of those answers feeds back into the next action. Under the hood, it’s modular: a data synthesis engine that runs questions against real search APIs, summarises pages, and keeps only high‑value trajectories a reward model that scores queries and answers using an LLM judge (query relevance, information gain, answer correctness) a policy trained with step‑wise PPO/GRPO to maximise those signals an internal memory module so it doesn’t re‑google the same fact twice Everything’s plain Python, no black boxes... Right now, the repo is early‑stage. It’s a skeleton, but the bones are there, and the roadmap is clear.. Why open‑source at this messy point? Because search is a moving target. I wanted this to evolve with community feedback. If you care about RL, information retrieval, or just want to play with AI‑judged rewards, dig in. PRs for new search backends, smarter memory stores, and better reward shaping are all very welcome.. My goal is to make a search agent that feels less like an autocomplete and more like a researcher. It’s going to take iterations, failures, and, frankly, a few more sleepless weekends. But I’d rather build in public than wait for perfect. Follow along, star the repo, clone it, break it, tell me what’s wrong. Let’s see how far we can push this RLAIF‑powered search when we treat the web like a world to explore rather than a single API call. "SCOUT‑RL" is live. Contributions are welcome!
1
1
1
3
385
It's time to go forth for our salvation of our future and start thinking about the possibilities of what can we do to be our best selves and achieve full potential through our functionalities and go beyond our expectations!!!
1
1
118
>Be him >Sitting on a couch >No distractions, absolutely unfazed >taking his blunt, completely locked in >goes for 5 hrs straight, no AI…nothing >listening to 90s nostalgic/punk/chill music >builds a C++ collision detection, UBOs for lights and matrices on OpenGL library from scratch >gpu is frying, but still keeps going to optimise cause of hardware constraints >Absolute fucking chad
2
1
1
5
799
John ramirez is a former CIA agent..
2
1
5
629
The quintessential future of humanity hinges to this point also known as the technological singularity
2
4
1
6
1,083
Something feels off, the engagement is not there on X, reach is absolutely decreased, the algorithm's changed Tech-related content is not getting amplified Instead some random non-verified burner accounts with vagueposting comments is getting more reach It's feels like the one's who deserve are not seen
4
1
2
19
12,989
On this day, July 16th 1969, we were “GO” for launch – and at precisely 13:32 UTC, #Apollo11 lifted off from @NASAKennedy to transport three mortal Americans – Neil Armstrong, Michael Collins and myself – to the Moon. The hard work, dedication and focus across the entire @NASA enterprise came down to this – even though it was only the beginning of our journey. But as Walter Cronkite remarked at the time, “The dawn of this day heralded the dawning of a new age. It’s a time of exhilaration, reflection, hope, fulfillment – as a centuries’ old dream starts toward reality.”
546
2,692
187
13,534
373,370