← Back to leaderboard
astridtgtz

astro

Verifiedactive⚠ low-confidence
@astridtgtz

building with AI, not just talking about it | content strategist

RosterAI Agents & Productivity ToolsAI Content Creation & Video Gen
65,586 followers · 666 following · Last polled 3h ago
Perf. Score
64.9
Rank #109
Value Score
86.3
#35 of 273 priced
Median views
21K
p25 21K · trailing 7d
Followers
66K
+0.1% 7d
ER (impr.)
0.75%
eng ÷ impressions
ER (foll.)
0.24%
eng ÷ followers
Est. CPM
$1.94
QT-rate basis
WoW views
+35.8%
▲ rising

Rates & value · est. from 21K median organic views

underpriced
Quote tweetbasis
$40
$1.94 / 1K views
Post
—
no CPM
Thread
—
no CPM
Retweet
—
no CPM
Value Score 86.3 · rank #35 of 273 priced$258.06 / 1K engagementsrate age unknown (pre-audit import)

Trends

Follower growth
Median views / day
Engagement rate
View distribution · trailing-7d posts

Each blue dot is one organic post. The Performance Score ranks by the median (21K), not the mean (21K) — so a single viral post doesn't inflate a creator's reach. The p25 floor (21K) is the level 75% of posts clear.

Not enough in-window posts yet to show a distribution.

Recent posts · ranked by views

PostPostedViewsLikesRepostsRepliesQuotesBookmarksER
YOUR AGENT DOESN'T NEED AN ESSAY. IT NEEDS A YES OR NO Most of what an agent does is decide: → Which tool to call → Which route to take → Does this pass the policy check Every one of those usually waits on a model writing paragraphs. Drex 1.5 skips the text. Send a state and typed questions, get a probability for every option in one forward pass. Fastest decision model on @OpenRouter right now. Built by @NaceAI.frozen12d ago46K118110990.48%
🚨 DO YOU REALIZE WHY ALMOST NOBODY EVALUATES THEIR LLM OUTPUTS IN PRODUCTION? it's not that teams don't care. it's the bill. grading an output properly means running a frontier model as a judge on every response. so you sample 2% of traffic, call it observability, and never see the other 98%. the alternative was a small classifier. fast and cheap, but it only recognizes the behaviors it was trained on, and the failures you actually care about are the ones you didn't predict. Respan just shipped Span-01 for exactly that gap: a hyper-parallel reasoning classifier built for unseen behaviors, trained with RLAIF. → #1 on the Behavior Benchmark → 2x cheaper and 18% better than Jev → 700x cheaper and 4% better than GPT-6 Luna → Span-01 Lite is free, and still scores above Jev nobody debugs what they never measured. nobody samples 2% and calls it coverage. when judging an output costs 700x less, you stop sampling and start scoring everything.
frozen
27d ago
24K
36
1
7
0
1
0.19%
one DGX card, a model built for efficient inference , Ornith-1.5-122B belongs on your shortlist2d ago21K15130010.75%
i imported an old powerpoint into the new Gamma: a dense table i made to study how the spanish government is organized.formatting intact. then i gave the agent one instruction: "turn this into something i can actually review before an exam."it asked me a couple of questions before building anything. then it rebuilt the deck and i exported it back to powerpoint. Gamma 5 is out today @GammaAppfrozen16d ago20K710000.04%
"Meta ads are broken" sounds like bait But the point underneath it is pretty simple: right now you're paying just to find out which creative actually works. Relay flips that order. You test a bunch of hooks, personas, formats and messages organically first, for way less money. Approved posts go out through a network of 50,000 human creators or your own TikTok account, and every result gets tracked. Then you take what you learned over to Meta, TikTok or whatever paid channel you're running, and you own the content. So by the time you're spending budget, you've already seen what works and you're not guessing anymore. Very neat idea and execution by Relayfrozen21d ago19K1030010.07%
Got early access to Griffin from @tavus. I told it: "I'm practicing a speech. Interrupt me if I ramble." Thirty seconds in, it cut me off mid-sentence and told me exactly what I'd left out. Every AI I've used waits for its turn. This one doesn't.frozen21d ago18K1300000.07%
still the best breakdown of enterprise AI on this app:frozen14d ago15K42126020.40%
Confident, completely made-up traffic numbers. that's what I kept getting from my agent. not its fault, it had nowhere to look. with SpringBrand connected it's pulling real numbers now. took me way too long to see that was the gapfrozen24d ago12K10911010.94%
this just killed 1000 SaaS startups https://t.co/Ff4RGjp6uxfrozen7d ago7.7K31100040.58%
COMPUTER USE HAS BEEN STUCK ON SCREENSHOTS. Pine Computer takes a different route: → a cloud computer built for AI, not for humans → reads websites as structure, not pixels → one SDK: you send a job, you get the job back done The numbers: → 78.3% on SaaS-Bench v1.1, the top checkpoint score on the publisher's board → ~1/20 the token cost of GPT-5.6 Sol + Codex → 1/26 the token cost of Opus 5 + Claude Code Agents don't need better eyes. They need a computer that speaks their language. Private beta at https://t.co/HByDTah3Qg #PineComputer #ComputerUse #DeveloperToolsfrozen13d ago6.4K1017010.30%
WHY DID AI GET CHEAP WHILE EVERYTHING ELSE GOT EXPENSIVE? That's the question nobody asked while we were busy benchmarking models. The markup survives on: → Information you don't have → Comparisons you won't make → Contracts nobody reads → A middleman between you and the price Aonic Club is one answer: point the intelligence at the cost of living instead of the demos.frozen9d ago962500000.52%
A VIDEO MODEL PREDICTS PIXELS. A WORLD MODEL PREDICTS CONSEQUENCES Same output on your screen. Completely different thing underneath. A video model: → Learns what footage looks like → Physics as a side effect → You watch the result A world model: → Learns how objects behave → Physics as the objective → You act inside it Odyssey-3 just set the state of the art on Physics-IQ. That's the benchmark that decides whether a robot can trust what it simulated. 🔖 Save this.frozen14d ago808600010.87%
GROK BOT JUST GOT AN OPEN-SOURCE ALTERNATIVE Same idea. Bots in a group chat with you and your friends. Different rules. Grok Bot: → Runs in the cloud → Its own bots → You join their platform @alook_ai: → Runs on your machine → Your own agents → Open source, self-host it Claude Code, Codex, Cursor, OpenCode, Pi. Whatever you already pay for. Most teams don't need new agents. They need a room where the ones they have can work together. 🔖 Save this.frozen14d ago556400010.90%