Building domain-expert AI agents across products, research & startups. Open-source tools, pet projects and field notes on context, evidence & judgment.

Dubai, UAE
Joined October 2009
jc_stack retweeted
I made an ad for GPT Image 2.5, full prompt below. It's yours now. Put your own character in @[image1], your own poster in @[image2], change the city, change what she turns into. Quote this with your version, I want to see what you all make of it. Seedance 2.5 on @higgsfield. Seedance 2.5 Prompt: [GLOBAL] 30-second commercial. Photoreal live-action cinema texture, digital cinema camera, 35mm and 24mm lenses, shallow depth of field, real overcast rain-light, low-contrast cool grade, light film grain. 16:9. The spine of the film is one continuous over-the-shoulder tracking shot; all transitions are hidden inside motion blur and the crystallization effect. [CHARACTER ANCHOR] @[image1] = the protagonist, a Japanese woman in her early twenties with a Tokyo street-fashion look: bleached pale-blonde wolf-cut shag with dark roots and a single mint-green streak in the under-layer; clean Japanese makeup — thin brows, pearlescent eyeshadow, nude lip; an oversized charcoal coach jacket with an original graphic print on the back, a long white tee underneath, baggy black cargo pants, chunky platform sneakers, layered silver jewelry, wired over-ear headphones around her neck, a small crossbody bag. Her face, hair and wardrobe stay 100% identical from 0s to 28s — never altered, never touched by any "generation" effect. She is the only constant in the film until the final beat. Emotional arc: oblivious → noticing → confused → stunned → awestruck → transformed. [ARTWORK REFERENCE] @[image2] = the billboard artwork reference. Used ONLY for the flat graphic displayed on the bus-shelter digital billboard at 11–15s — including its complete typographic lockup ("Higgsfield × OpenAI" in small caps, "GPT IMAGE" in heavy white, the lime-green gradient "2.5", and the wide-tracked "SUNBURST" beneath), along with its layout, palette and composition. @[image2] must appear as a complete flat advertising image inside the billboard screen, carrying correct perspective, screen reflection and rain refraction across the glass surface. The woman pictured in @[image2] must NEVER be used for the protagonist's design, must NEVER leave the billboard screen, and must NEVER appear in the physical space of the scene. @[image1] and the woman in @[image2] are two entirely different people — never conflate them, never let one influence the other. [CORE VISUAL MOTIF — THE ONLY TRANSFORMATION LANGUAGE IN THE FILM] Anything being "regenerated" must follow the exact same three-stage process: ① the original object disintegrates from its edges into a cloud of tens of thousands of small translucent white crystalline beads — like bubble-film or shattered glass pearls, with real volume, specular highlights and real weight; ② the bead cloud hangs briefly and rapidly rearranges; ③ the new form grows outward from inside the cloud, the final layer of beads absorbing into its surface and vanishing. Do NOT use flash cuts, dissolves, particle-dispersal effects, or digital-glitch styling for any transformation. [SETTING] A Midtown Manhattan intersection after rain. Wet asphalt reflections, crosswalk stripes, yellow cabs, dense glass curtain walls and older stone buildings, neon and storefront signage, a fire hydrant, a newsstand, a bus-shelter digital billboard. Overcast diffused light throughout, no direct sun. 0–3s: Over-the-shoulder medium shot, following behind @[image1]. Her back is to camera as she stands on the crosswalk. To frame right, a 1970s green sedan in motion crystallizes entirely and reassembles as a black muscle car; a few frames later it crystallizes again and reassembles as a white wheel-less hover car. She hasn't registered it yet. 3–7s: The camera arcs around to her side. She starts looking around, the ends of her wolf cut swinging with the head turn. A businessman with a briefcase crystallizes mid-stride — the human-shaped bead cloud holds the walking pose for two full steps — and reassembles as a bare-chested barbarian warrior with feathered ornaments, continuing the same stride. Two or three more pedestrians crystallize in the background simultaneously, without pulling focus from her. 7–11s: Handheld tracking accelerates. More and more people on the street are replaced: a purple-skinned alien warrior woman, white-armored robot troopers, an original gold-suited superhero character (all original designs — no real-world recognizable characters, logos or trademarks). She walks faster, keeps turning to look back, her breathing quickening, one hand instinctively gripping her crossbody strap. The replacement frequency keeps rising, but the street's spatial relationships, lighting and wet reflections remain continuous throughout — no jump cuts. 11–15s: She stops in front of the bus shelter. The digital billboard inside cycles rapidly: a photoreal movie poster → a hand-painted concept illustration → a fashion editorial — each refresh executed with the same crystalline bead reassembly. It finally settles on the complete @[image2] artwork filling the entire screen, its typography and palette exactly as in @[image2]. She raises a hand and points at the screen, mouth open, unable to speak. 15–18s: The camera tilts down to her feet. The sidewalk paving crystallizes outward from where she stands in a perfect circle, white beads spreading three meters along the joints between slabs, the manhole cover and the standing water turning crystalline with it. She looks down. The ground gives out beneath her platform sneakers. 18–21s: Wide 24mm. The ground collapses and she falls straight down. The camera falls with her looking up — the entire city contracts above her into a shrinking square opening, the buildings on either side peeling away, tumbling and receding upward like building blocks, the sky turning to deep blue-black void. She throws her arms out, her hair, jacket hem and bag strap lifted by the updraft. 21–24s: During the fall the visual style jumps repeatedly, each jump bridged by one full-frame sweep of crystalline beads. In order: ① rich 1980s film stock — high saturation, heavy halation, heavy grain; ② black-and-white halftone manga panel — the character rendered in ink linework and dot screens against dense radial speed lines; ③ a neon purple-blue digital vortex with long light streaks stretching backward at speed; ④ near-total darkness with extreme motion blur. Her facial structure, hair and wardrobe must remain recognizably the same person in every one of these styles. 24–27s: She lands flat on her back on a smooth white floor. A wide lens pushes in slowly and tilts up to reveal the environment: an infinitely extending pure-white showroom/hangar with a gridded softbox ceiling. Hundreds of black circular pedestals cover the floor, each holding a character statue or a parked vehicle — original-design warriors, mechs, robots, sci-fi troopers, plus large military aircraft and helicopters. A small line of floating white label text sits in front of each pedestal: "FANTASY CHARACTER", "VEHICLE", "MECH" — set at a small point size, present purely as environmental information. 27–28s: She sits up and pushes herself to her feet, turning slowly to take in the hall, eyes widening in awe. The camera settles into a full-body medium-wide, centered, the statue array receding out of focus behind her. 28–30s [THE FINAL TRANSFORMATION]: The instant she stands fully upright, the crystalline beads rise from the floor around her feet and sweep up her body in one continuous wave — and for the first time in the film, SHE is the thing being regenerated. Her streetwear disintegrates into the bead cloud and a magical-girl battle uniform grows outward in its place, following the exact same three-stage motif, generating in a readable order from the torso outward: a white high-collar fitted bodice → a mint-green sailor collar with three silver stripes → a deep mint pleated skirt with silver trim, its pleats opening one by one → long white gloves to the upper arm with mint piping → white knee-high boots with a mid heel and mint cuffs → layered silver jewelry reforming as a silver star brooch at her chest → a large mint-green bow at her back with long ribbons falling to the backs of her knees → a fine silver tiara condensing across her forehead. Her face, hairstyle, hair color, the mint streak and the headphones around her neck remain completely unchanged — only the clothing is regenerated. The last beads absorb into the fabric. She looks down at her own hands, then slowly raises her head, the ribbons still settling behind her, and looks straight past camera with a small, dawning smile. The film cuts on that frame. No end card, no title, no tagline, no subtitle of any kind. [SOUND] Rain and wet-road traffic, urban ambient bed; high-frequency glass/ice shatter-and-rearrange sounds on every crystallization; a short bright locking tone as each reassembly completes; electronic refresh tones from the billboard; low-frequency collapse as the ground gives way; wind through the fall, with a distinct timbre shift for each of the four style jumps (film hiss / paper-and-ink texture / synth sweep / silence); one muffled impact on landing; near-total quiet inside the hall with only faint HVAC room tone; then, over the final transformation, an ascending shimmer of silk and crystal, each garment piece seating with a crisp click, the bow tying with a single sweep of ribbon, and one clear bell tone as the tiara locks in. One escalating electronic music bed ties it together, landing its final hit on her raised head. [HARD RULES] @[image1]'s face, hairstyle, hair color and headphones are absolutely unchanged across all 30 seconds. Her wardrobe is unchanged from 0s to 28s and may be regenerated ONLY in the 28–30s final beat — the transformation affects clothing only and must never alter her face, hair or identity. All other replacements happen only to the people, vehicles, ground and environment around her. The film may use only the single "crystalline bead disintegration → rearrangement → growth" transformation language, including for the final costume change; no other transition effect may be mixed in. From 0–18s the film must hold the logic of one spatially continuous tracking shot — the street's architecture, lighting and wet reflections must never jump. Every character and vehicle in the hall must be an original design; no real-world recognizable characters, brand marks, trademarks or identifiable film properties may appear, and the final magical-girl uniform must be an original design, not any existing anime character's costume. The only brand copy permitted anywhere in the film is the @[image2] artwork on the billboard screen at 11–15s, existing solely as a flat image inside that screen and never repeated elsewhere. No end card, no title card, no subtitles, no watermark, no overlaid text anywhere in the film. Avoid: cartoon rendering, game-CG feel, plastic materials, full-frame bloom, digital-glitch transitions, weightless floating, incorrect finger structure, garbled text, revealing or suggestive posing or camera angles.
🤝 Paid partnership · 🤖 Made with AI
34
55
4
530
64,699
jc_stack retweeted
I still think about this a lot
233
897
194
15,180
1,948,728
RT @DrJStrategy: Crude Oil Fibonacci Retracements. $70 is the target Have a nice day.
2
13
jc_stack retweeted
Replying to @KayRooze
may just be the two religions going up against each other
10
2
124
4,926
The solo builder edge: speed. No meetings. No approvals. No committees. You can pivot in a day. Ship in a night. Test ideas before others finish planning. Move fast. That’s your superpower.
5
4
39
5,172
I love when Claude tells me something will take 6 weeks. No dude, we are doing it this afternoon.
169
615
84
9,916
462,951
jc_stack retweeted
Why I think $STRATOS is cool ​ Let me preface this by saying that as a partner in the Virtuals Partner Network program, my strategy is to bring in unique, value-adding product into @virtuals_io ecosystem ​ Why? ​ @BasisOS opened my eyes to the possibility of bringing robust Defi protocols with an AI roadmap to Virtuals—Defi protocol that offers real value, real yields, not an entertainment/conversational bot, but a product that offers real yields. ​ On Virtuals side, there's an ongoing need for improvement in liquidity, need for better trading environment, need for fundamentals that Defi legos can be built upon. ​ That first lego piece is Stratos by @IthacaProtocol ​ Stratos is a swarm of options trading agents that have been profitable and live for more than a month. ​ Defi options certainly has its own challenge as a sector—from liquidity fragmentation and capital inefficiency to being a highly complex product w/ weak demand dynamics BUT STILL options are very useful ​ - Back in 2008, everything was crashing massively. Some hedge funds & instos long put options on S&P500. While portfolio dropped 50%+, their protective puts limited further drawdown, allowing them to outperform peers. ​ - Trading options gives you the flexibility to bet on price direction OR the volatility itself—specifically the gap between implied volatility (IV) and realized volatility (RV) ​ (Think of IV as forecast and RV as the actual weather) ​ - Market tend to over-react ahead of known events (e.g. FOMC, earnings report, CPI) ➔ IV rises sharply. ​ Often times, actual price move is smaller than expected. Traders can sell options before the event, expecting IV to collapse afterwards. ​ - This predictable patterns around scheduled events create profitable vol trades for traders ​ The above is a simplified example but you get the idea ​ Stratos does this for you with more complex strategies that take advantage of the IV & RV mispricing. Each agent within the swarm have their own strategy and personality—some more profitable than the other, some are highly degen but others may be conservative. ​ How does AI come into play? AI/ML is used for pattern recognition, regime classification, and signal refinement ​ Won't go full on technical here. If you're interested, the team has a technical piece you can dive into. ​ What Stratos does is - It watches the market constantly, observing prices & volatility - It gives two types of scores (i) Spot Trend (ii) Volatility Regime - It uses AI to detect pattern ➔ AI tools like Hidden Markov Models (HMMs) help it recognize when the market is shifting—like a weather radar noticing a storm forming - It adapts strategy based on the environment e.g. calm market + uptrend maybe sell puts, volatility too cheap? maybe buy a straddle - It adjusts for crowd behavior ➔ depending on whether traders are all thinking the same way, the swarm acts more cautiously or aggressively ​ How Does Stratos Make Money? ​ The swarm charges 20% performance fee with 50% of this goes to $STRATOS stakers in the form of real yields ​ So far, it's managing ~$24k in portfolio value, generating ~20% APY. The team aims to increase the agents bankroll further to make this more exciting ​ Bottom Line ​ When I look around Virtuals ecosystem, there are only a handful number of teams that provide quality product. Many Genesis Launches launch Day 1 with only a roadmap, Stratos agents have been live running for >1 month with profitable strategies. ​ The Ithaca team is one of the top Defi options w/ extensive experience. Ithaca infra makes options trading a lot more efficient, serving as a great foundation for options applications & options agents to deploy on. ​ The team also has cool options applications that can further enhance ACP growth as well as part of their roadmap. ​ ———————————— Personal Note ​ Thanks for reading ser. Disclaimer: I'm a Virtuals partner supporting Stratos so I might be a little biased here. Do take everything with a grain of salt. This is not a financial advise, make sure to do your own research before pledging the points or making any decision.
155
24
22
234
30,985
Incredibly exciting AI language in the @HouseCommerce budget reconciliation text. Includes provisions to modernize government and a moratorium on state-level AI legislation, preventing US innovators from being crushed under an avalanche of regulation.
23
31
26
178
231,697
⭐️@RLanceMartin is back on the podium to show how to build a long-running "ambient" agent that responds to emails and updates your calendar! Automating your inbox saves time but you DEFINITELY want final review before sending responses. Human-in-the-loop UX with LangGraph is the answer! Stay tuned - more to come here after a break 👀
1
7
1
29
11,040
jc_stack retweeted
If you believe in the future, build it. “Technology is the glory of human ambition and achievement, the spearhead of progress, and the realization of our potential.” - @pmarca
31
66
10
524
98,120
Cursor agent's prompt and tool use is so good and pragmatic, it shows how far you can go with just 1300 tokens, 10 tools (and "the world's best IDE"). No AGI needed. Prompt and tools 👇
1
7
3
55
12,222
MORALIS HAS NO TOKEN DON'T FALL FOR SCAMS
11
7
79
8,860
LLMs have made learning a lot more fun for me. It hasn't made me lazier (yet) as I might've expected. I find myself being more curious, and learning a lot more per day. This is a bit surprising to me, and awesome. Same goes for programming. Removing all the code I generate with LLMs from consideration (which is a lot), the amount of code I write on top of that has also increased. Again, to me, this is surprising. I think it's because I'm more energized and excited by the overall productivity boost. Anyway, we live in exciting times!
393
369
89
5,155
1,039,876
✨ Observe and Evaluate Multi-Modal Agents with LangSmith 🖼️ LangSmith now supports images, PDFs, and audio files across the playground, annotation queues, and datasets — making it easier than ever to build and evaluate multimodal applications. 📹 Watch how we evaluate a receipt-extracting agent in action: youtu.be/-VOnEpk3uWo 📄 Check out the docs to get started: docs.smith.langchain.com/eva…
4
38
4
174
13,175
I'm still not actually worried about this? I'm not anti-human I just remember learning about carrying capacity. Not in a Malthusian sense. Just plain biology. "Finite habitat means finite population" Furthermore, the massive social contract shift (namely women's rights) necessarily has changed social dynamics. Am I alone here?
Something seismic is happening across the globe: birth rates are collapsing to levels not seen in modern history. This isn’t just a blip-it's a profound shift affecting nearly every developed nation. But why is this happening, and what does it mean? Let’s break it down:
108
10
2
195
47,746
There have been 3 core UX metaphors for computing. We are about to see a fourth one -- the AI computer. A 🧵 1. In 1981 MS-DOS was released. The UX was invoking terminal commands like "dir"
34
15
4
121
35,773
🛠️ Sidekick: Trellix's Engineering Assistant Trellix built Sidekick, transforming engineering workflows by leveraging LangChain's product suite (LangChain, LangGraph, LangSmith) to create an AI-powered development assistant. Read their story 📖 blog.langchain.dev/customers…
2
7
25
7,830
converting unstructured data (tweets) into structured data so i can turn it back into unstructured data (daily newsletter)
look at my beautiful boy
18
20
247
19,255