TradeFox (@tradefoxai) header
TradeFox (@tradefoxai) profile picture

Prediction market aggregator and prime brokerage. Backed by @alliance and @CMT_Digital.

New York
Joined February 2025
We’ve added 100s of new smart money wallets, categorized by strategy and edge: - Trading bots - Bonding Soon Whales - High Win Rate Wizards - Hyped High Volume Market Makers - Manual Movers - Wall Street Wizards, - Insiders, Sports, Politics, Weather, etc Copy trade any of them in one click.
14
17
3
63
42,871
TradeFox retweeted
Human brains are just not suited to comprehend exponentials Updated $ANT model: - 2026E: $81B ARR Oct'26 +14x YoY -$14B FCF - 2027E: $201B ARR + 149% YoY at -$62B FCF - 2028E: $363B ARR +80% YoY at -$33.8B FCF - 2029E: $562B ARR +55% YoY at +12B FCF - 2030E: $800B ARR + 42% YoY at +$90 FCF
UPDATED ANTHROPIC MODEL: We revise down our Anthropic 2030 ARR to $800B ARR down 20% given @SemiAnalysis_ revision on AI capex. View: Anthropic will go public this week and I believe $ANT trades end of week at $2.8T (2x from IPO price) with $800B ARR '30 and $1.6B total revenue from 2027-2030 that offsets the $900B of compute already committed. Implies 9.2x '27 revenue / 4.6x '28 revs. Thesis: Anthropic is the #1 winner of enterprise AI and agents are still early on the adoption curve (<1% business adoption globally). Growth mode makes sense. We believe Anthropic gets to $800B ARR in '30 w/ 50% marketshare for enterprise token spend and with Anthropic Direct, can get to 13% FCF margin and 28% EBIT margins in 2030. Context: Biggest threat to Anthropic was open source tokens being significantly cheaper than SOTA models e.g Fable / Mythos and enterprises shifting spend. But... with Jensen out at $NVDA = BAD for open source since less chips to train / distil = GOOD for $ANT marketshare. Here's our updated model: - 2026E: $81B ARR Oct'26 +14x YoY -$14B FCF - 2027E: $201B ARR + 149% YoY at -$62B FCF - 2028E: $363B ARR +80% YoY at -$33.8B FCF - 2029E: $562B ARR +55% YoY at +12B FCF - 2030E: $800B ARR + 42% YoY at +$90 FCF Here are our base case assumptions: 1. REVENUE: $1.8T of Enterprise spend in 2030 up 10x from 2026 Estimate of '26 $180B ARR of AI Enterprise spend from @MenloVentures > Fortune 5000 go from 0.1% token spend of OPEX to 10% and agent adoption goes from <1% to 10%+ NOTE: our base case doesn't account for $ANT Ode with Anthropic (take rate on services), drug discovery / life sciences, other areas ... we just project out token spend to replace / augment knowledge work. Can easily add another $500B ARR to 2030. 2. REVENUE: ANT owns 50% of enterprise spend in AI in 2030 -- we think open source is no longer a strong threat. > OpenAI is the biggest credible threat at 30% market share (Open Source we have as <10%) 3. COST: Anthropic Compute Assumptions - $70B ARR per inference GW due to token efficiency - $17B Cost per GW driven by Anthropic Direct and owning / leasing data centers directly - less $ and faster time to launch - 25 GW secured by EOY 2030 requiring another 15 GW in addition to $900B already committed. KEY: we believe $ANT will generate $1.6B of revenue next 4 years which pays back the $900B of compute Anthropic has for leases that run through 2036.... Get the full model here: artemis.ai/company/ANTH?tab=…
4
4
14
883
TradeFox retweeted
i stand by this prediction the hero's journey is not without its trials and tribulations. dragons must be slain, sirens resisted, and the underworld braved. only then can the hero return home. and the return is imminent.
Jensen should be TIME Person of the Year and it's not even close
2
4
9
473
TradeFox retweeted
discovered an outreach hack that saved me 100 hours linkedin doesn't expose itself to agents, obviously, but here's what you can do instead 1. go to settings & privacy -> data -> get a copy of your data and select "download the larger data" 2. you'll get a zip in your email 3. unzip the folder and upload the "connections csv" to claude alongside a list of the target companies you're trying to get in touch with it just identified 133 target connections for me linkedin is legally required to let you download your own data just saved about 100 hours you're welcome
2
4
8
615
TradeFox retweeted
the gagosian of datasets the models are getting really good at generalizing, so most private datasets aren't actually that valuable. the datasets need to be outliers to really improve the models and therefore be valuable. so it's almost like you want to build an old-school antiques or art dealership firm and network. you want to build the gagosian of datasets
2
4
7
668
TradeFox retweeted
tech ate media, finance, entertainment, retail, automobiles, aerospace, and defense. now two institutions remain: higher education and organized religion. these are the hardest to conquer because engineering excellence is insufficient. you need the sincerest understanding of what keeps society in precarious balance: status rationed through scarcity and fear quieted through faith.
4
1
8
794
TradeFox retweeted
this goes hard if u don't understand the diff b/w dual-class super voting shares at 48% ownership and single class shares at 3.6%
2
4
9
551
TradeFox retweeted
this is spot on. earlier this year i had to let go of our front-end engineers. not a decision i took lightly, but opus + technical designer + backend was simply more efficient. here's what changed from where i sit: 1. front-end was mostly a translation job. designer makes it in figma, FE engineer turns it into components. models now do that translation really well, especially on standard stacks like react + tailwind where there's a mountain of training data. 2. handoffs were too costly. design to FE to backend meant two handoffs and fidelity lost at each one. a designer who can prompt and ship directly cuts one out. fewer syncs, fewer "that's not what i designed" msgs. 3. the evalset for UI is clear as day. you look at it and know if it's right, so the AI iteration loop is fast and cheap. backend mistakes stay invisible until they're expensive (data, security, scale), which is why humans still sit there. notice security, platform and backend are the roles growing in this chart. 4. the bottleneck moved to the edges. the scarce skills now are taste (knowing what good looks like) and systems (making sure the thing underneath doesn't break). pure implementation sat in the middle and got squeezed from both sides. if you're front-end today, the people i see thriving have moved toward design engineering or full-stack product work.
Engineering is changing: FDE is ascendant while mobile and front-end roles are disappearing. Open mobile and front-end engineers jobs have been nearly halved in the past year!
35
20
7
348
47,976
TradeFox retweeted
new data brokerage motion I’m seeing: ai services firms close mid-six-figure to low-seven-figure enterprise engagements for workflow automation (sales ops, back office, support, etc) by positioning downstream data licensing deals with frontier labs as an additional payback mechanism on top of the workflow roi. the automation generates proprietary task trajectories, labeled edge cases, and domain-specific eval sets. positive-sum across the stack: 1. enterprises: services spend is offset by licensing revenue, so it’s close to cash-neutral, and they keep the efficiency gains 2. services firms: high-margin implementation fees plus a cut of the data deal 3. labs: proprietary, domain-specific training data plus increased inference volume from the deployed workflows win, win, win
3
4
10
791
TradeFox retweeted
Will fable 5.2 and mythos 5.2 demonstrate that Sutton's bitter lesson will hold in robotics?
50%Yes
50%No
8 votes • Final results
3
4
7
843
TradeFox retweeted
banks’s mindset of abundance is pervasive in the Culture and is extremely useful for us, as humanity is on the precipice of an AGI-driven cornucopia
In Iain M. Banks's books, the Culture solves envy—a crux of humanity—by letting every citizen experience what they have not yet via tremendous compute (a world where citizens have no token limit). Citizens quickly realize envy is pointless, as different life experiences mix good and bad; the grass of others is not any greener.
2
4
6
615
TradeFox retweeted
this aged pretty well if i might say so myself
The neolab trade maps to alt L1s: high-beta proxies on the dominant narrative, rerating faster than the underlying because supply of investable exposure is constrained. Bare metal maps to BTC miners circa 2019-21: operating leverage on a commodity input, providing convexity on the way up. The key is knowing which inning we're in.
1
4
8
645
TradeFox retweeted
🤯 RSI is already here -- it's not just very evenly distributed
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math. rikiyat.github.io/blog/weight-sp… github.com/RikiyaT/lm_und… huggingface.co/Qwen/Qwen3.5-0… huggingface.co/RikiyaT/Qwen3.…
1
4
8
610
TradeFox retweeted
me: have you tried selling your datasets to generate cash? him: how would that work? me: you have 66m datapoints. convert your raw unstructured data into rl tasks: user input -> agentic reasoning traces -> output -> evals him: why would anyone want that? me: labs will pay millions to train their models on this him: you can't model ball knowledge me: smh
8
6
91
7,947
TradeFox retweeted
2000: dot com 2008: subprime 2020: covid 2026: ai cyber
1
4
1
7
729
TradeFox retweeted
if this allegation is true, here’s how the mechanism could work, step by step: 1. broker sells training data to ai labs. - codebases, expert demonstrations, synthetic examples, records of people completing tasks. - the labs buy these inputs to improve specific capabilities. 2. those transactions reveal what the labs are working on. - what data they need. how much they’re buying. which capabilities they’re prioritizing. their deadlines. - for example, a large order for accounting workflows could suggest a push into accounting automation. 3. developments at the lab affect publicly traded companies. - the lab itself doesn’t need to be public. - a new agent could threaten a software company’s revenue. a large data contract could boost a supplier’s earnings. a cheaper model could change expectations for spending on compute. - the broker connects what it knows about the lab to the companies whose shares could move. 4. the firm takes positions before the information becomes public. - when the news becomes public, the market reprices those companies. if the reaction matches the firm’s expectations, it profits. - the profits get presented as evidence of exceptional trading models.
This evening I sat across from the CEO of a so-called trading neo lab with over $1B val. He told me their models—built on best-in-class RL environments and online learning algorithms—have produced exceptional pnl and sharpe ratios. I later learned that the company is, in fact, a data broker LARPing as a frontier lab. Its actual edge is trading stocks and options on insider information of model releases and synthetic-data acquisitions. It appears I have once again materially underestimated the absurdity of this industry.
2
7
1
42
4,177
TradeFox retweeted
data brokerage businesses seem like a flash in the pan to external observers, but these businesses will have greater terminal value than you'd think. why? 99.9% of the world's data is still private. the labs are locked in an arms race to acquire it. - runbooks - crm records - jira histories - multi-tool desks - meeting room av - app internal logs - salesforce archives - help center articles - sales call recordings - slack channel history - support triage events - browser checkout sims - spreadsheet agent evals the list is endless. i agree the absurd margins that datasets command will not be sustained indefinitely as supply increases to meet the demand, but the demand is real and enduring companies are being built.
5
7
2
61
6,697
TradeFox retweeted
the roadster demo is yet another reminder that life expands or contracts in direct proportion to your level of courage
Turns out the future does have flying cars
5
12
723
TradeFox retweeted
6 months ago I asked if anyone was building an Etherscan / Arkham for LLM reasoning traces. Reading this makes me want to take that idea deeper: an explorer for the neurons and circuits inside the model. Click a neuron. See what activates it. Follow its connections. Test what happens when you turn it up or down. Nick’s argument is that this type of mechanistic interpretability may be much more doable than we thought. His early OpenAI work found only ~2% of neurons interpretable. With better tools and modern models, he now reports that the vast majority are understandable, including in models with trillions of parameters. And every neuron you understand gives you clues about the ones connected to it. Basically, the map gets easier to fill in as you go. AI agents can help do this at scale: propose explanations, run experiments, and trace the circuits. If this holds up, the same progress making AI more powerful could help us understand and steer it. This is a big deal for alignment.
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works already-clean.goodfire.com/
4
11
819
TradeFox retweeted
my love language is increasingly a rawdogged google doc or spreadsheet written with zero ai arxiv.org/html/2510.24011v1
5
6
593
TradeFox retweeted
i probably shouldn’t share this alpha openly but @boardyai has outperformed linkedin, twitter, and every recruiter i’ve ever used. combined. how it works: → i tell it exactly who i’m looking for → it intros me to matches over email (double opt-in) → checks both calendars and books the call → sends reminders + follow-ups → records, transcribes, and summarizes every interview → puts every candidate in one doc so deciding is easy not going back.
2
4
6
644