@0xRemington once you stop using Claude like google everything changes
English
RetroRanger Anonymous
3.4K posts
RetroRanger Anonymous
@RetroCoderX
MemeLord 🧠 Hanteln heben 💪
Nuremberg Katılım Ağustos 2021
183 Takip Edilen41 Takipçiler
If you only watch one AI lecture this month, make it this one from Stanford.
Not because it'll teach you better prompts.
Because it'll change how you think about AI.
Most people use ChatGPT and Claude like smarter search engines.
Ask a question.
Read the answer.
Repeat.
The lecture explains why that's only scratching the surface.
Once you understand how these models actually reason, remember context, and respond to structure...
You stop writing prompts.
You start designing systems.
That shift alone changes everything.
I took the biggest ideas from talks like this—and months of building with Claude—and turned them into a practical guide you can apply immediately.
Not theory.
Real workflows you can start using today.
Remington@0xRemington
English
Own the Agent learning LOOP. Rent the model.
Private evals, traces, memory is how is compounds.
Satya Nadella@satyanadella
English
@mikekalilmfg NASA Valkyrie to shipyard welding is one hell of a career pivot
English
RetroRanger Anonymous retweetledi
The Texas startup Persona AI is training its heavy-duty humanoid robot for welding work at shipyards.
The Houston-based robotics firm just shared footage it describes as a real welding test inside an industrial fabrication shop. The Gen 1 humanoid is guided through teleoperation to collect real-world welding data to train its AI brain for future autonomous work.
Persona AI says it's secured deployments with HD Hyundai, one of the world's largest shipbuilders, along with a steel-fabrication pilot in Louisiana. It's working with Under Armour on protective performance gear for the harsh conditions.
The work Persona AI is targeting requires mobility and flexibility that fixed welding robots on production lines can't provide. It's scaling production with at least $27 million in funding raised to date.
Persona was co-founded in 2024 by Nic Radford, the former lead of NASA's Valkyrie space humanoid program, and former Figure AI CTO Jerry Pratt. The company is also adapting dexterous hand technology from NASA’s Robonaut 2 for its humanoid worker.
English
CLAUDE CODE + GOOGLE STITCH = A $10,000 WEBSITE. FOR FREE?
Stitch generates the UI.
Claude Code reads files like design.md and turns the mockup into a working project.
The workflow:
> describe the site in Stitch
> generate the design
> open it in Claude Code
> build the full site
> edit it with plain English
Need another section, different copy, new colors or animations? Describe the change and Claude updates the code.
Stitch handles the interface. Claude handles the implementation.
One prompt creates the design. The next turns it into something you can actually ship.
Adea@Adea0x
English
@ziv_ravid they're basically paying people to use it at this point
English
I wonder whether, in retrospect, Anthropic thinks it made a mistake in developing such a large model. It made more sense in February, when there was no model near its performance level, but now, their only option is to keep subsidizing it through the paid plan
Claude@claudeai
We're extending Claude Fable 5 access on all paid plans, as well as keeping Claude Code’s weekly rate limits 50% higher, through July 19.
English
@argofowl @Suheil7020 as a german dev 'works but needs to be better' is painfully accurate
English
@Suheil7020 i agree, coming soon, german already works - but needs to be better
English
introducing namethatui.com
a dictionary for ui things you can see but can't name
made it because i'm primarily a designer, and my biggest resistance was always knowing what things are called when prompting my agents
it learns as people use it: every search teaches the site new words, and the built-in pocket dictionary grows with it
give it a try and let me know what you think
can't find something? dm me and i'll add it
i want this to be the lowest resistance resource you have
you can just build things
English
@Suheil7020 @argofowl Banger. other languages would be clutch tho
English
@argofowl the amount of times I've told a coding agent 'make it look like the thing with the dots' is embarrassing
English
@DAIEvolutionHub Qwen 14B beating o3 at email retrieval is wild
English
LLM fine-tuning techniques I'd learn if I were to customize them:
Bookmark this.
1. LoRA
2. QLoRA
3. Prefix Tuning
4. Adapter Tuning
5. Instruction Tuning
6. P-Tuning
7. BitFit
8. Soft Prompts
9. RLHF
10. RLAIF
11. DPO (Direct Preference Optimization)
12. GRPO (Group Relative Policy Optimization)
13. RLAIF (RL with AI Feedback)
14. Multi-Task Fine-Tuning
15. Federated Fine-Tuning
Since we're talking about fine-tuning, I wrote a full breakdown on fine-tuning LLMs with RL in 2026. Including how to skip manual reward engineering with automatic LLM-graded rewards.
And this is done using a 100% open-source solution: github.com/openpipe/art
English
@0xOrionVega cool rig but 'the rig replies' is doing a lot of heavy lifting here
English
One hour to build your first agent. Eighteen months later you've fired three teams and your revenue is up.
Four RTX 3090s. 96GB of VRAM stacked in one rig. Dual Xeon CPUs on a server board - two sockets running in parallel. 16 sticks of DDR4 lined up like a wall, 256GB feeding both processors. Riser cables so the cards can breathe outside the case.
This box runs his company now.
The four SDRs he used to pay $7k/mo each? Gone. The rig does the outreach. Three-person content team on retainer at $12k/mo? Gone. The rig writes. Two VAs handling support tickets around the clock, another $6k/mo? Gone. The rig replies.
That's $46k a month in payroll he cut. Revenue didn't drop - it climbed 34% in the first year. The machine doesn't sleep, doesn't call in sick, doesn't hand in notice. It works 24/7 and gets faster every time he tunes the prompts.
Under $6k in hardware, mostly used parts. Runs Llama 3.3 70B and Qwen 2.5 72B locally. Same specs rented on AWS would cost him $4-5k a month. This paid for itself in six weeks.
No API bill. No usage caps. No prompt logs on someone else's server. No payroll. No PTO. No turnover.
The no-code tutorial is the doorway. This is the room it leads to.
Start with the hour. See where it takes you.
DiKrass -X-@Di_Krass_
English
RetroRanger Anonymous retweetledi
the quant who managed $13 billion at Guggenheim:
"the better you get at backtesting, the less useful it is"
that was 2017, in a lecture on why machine learning funds fail
run 100 backtests on pure noise and the best one shows a sharpe near 2.5. so he demands 3. after 1,000 trials - 7. the bar never stops climbing
in 2026 an agent swarm runs those 100 in one night and calls itself a research team
bookmark & watch the talk, then read the statistics that catch a swarm lying to you - below
localminima@localminimaa
English
@THJohnG typisch. alle reden Gesellschaft, niemand baut die Pipeline
Deutsch
@lagerskoy 1 million agents and still can't simulate why my code works locally but breaks in prod
English
AI JUST TURNED PUBLIC OPINION INTO A SIMULATION ENGINE
This is not another chatbot demo. MiroFish shows how AI agents can simulate markets, people, events and information flows, then turn messy public behavior into something you can actually inspect
The workflow is the important part. You upload seed material like news, events or data, then the system builds a swarm of agents that interact, form opinions and expose patterns that would be invisible in a normal dashboard
The wildest moment is the scale. The video shows a simulation with 1,000,000 AI agents, then a report agent pulls out sentiment trends, key influencers and turning points instead of forcing a human to read the entire chaos manually
That is where this gets useful. AI stops being only a tool for answering questions and becomes a way to test how ideas, narratives, markets or communities might move before the real world reacts
The open-source angle makes it even more interesting. If systems like this become easier to run, every researcher, builder and analyst gets a small society-in-a-box for testing second-order effects before launching anything
localminima@localminimaa
English
this graph is hilarious, gemini is in the dumps
Sebastian Raschka@rasbt
Upon request, here's an updated version with Grok 4.5 and Meta's Muse Spark 1.1. Grok 4.5 seems to sit at the Pareto frontier. Good bang for the buck. (Also added harness info).
English
RetroRanger Anonymous retweetledi
So much progress in just a bit over a year. Here's what frontier first-person rollercoasters looked like back then: x.com/OfirPress/stat…
Vaibhav (VB) Srivastav@reach_vb
GPT 5.6 Sol is pretty GOATED at making games, it made an entire rollercoaster simulator with textures and assets included! It all started with a simple prompt in a /goal and with few artistic directions from my side, it's still not complete and there are some UI inconsistencies But you can go pretty far with a broad idea and feedback along the way! Goes to say, you're truly bounded by your own ambition Try it out yourself 🤗
English
RetroRanger Anonymous retweetledi
I thought it would be a nice small meet-up but it is a whole auditorium 🥶
Rudrank Riyam@rudrank
Claude New Delhi meetup! I am going first so I can enjoy the talks of other speakers 😆
English
@MuhammadShiha Connect the two and you basically got a policy simulation engine for an entire country. Not a twin, a crystal ball.
English
Three days ago I wrote about Abu Dhabi building a digital twin of its entire population.
Then Dubai launched one for the whole city.
Abu Dhabi's twin models people. Dubai's twin models the city. These aren't pilot projects, they're live infrastructure running government decisions in real time.
The question nobody's asking yet: what happens when you connect the two? At that point you're not modelling a road or a hospital. You're modelling what happens to people when you change them.
That's probably 3 years away. Maybe less knowing this place.
#UAE #DigitalPolicy #GovernmentAffairs
English
they mapped the world's power grid but my ISP still can't explain an outage
Token Gremlin@TokenGremlin
GPT-5.6 Sol has been working on Power Atlas for more than 16 hours straight without hitting a single rate limit. It is researching sources, reconciling conflicting data, verifying results and continuously expanding a global map of energy and digital infrastructure with almost no intervention from the user. That is the part worth paying attention to. Not the pretty map by itself, but the fact that the model can keep doing useful, structured work for this long without collapsing into loops, forgetting the objective or requiring constant babysitting. You can access the project and test it yourself here: power-atlas.sarvesh-kapre.chatgpt.site Cc: @_cyberhusky
English