+ (@noahphi999) header
+ (@noahphi999) profile picture

2x founder, ex ML eng, @Solana 2026 Privacy Hackathon winner 🛡️ Currently enjoying the singularity & the crash of the dollar prev @qualcomm,@tetsuocorp

New York, USA
Joined June 2023
Photo by + (@noahphi999)Photo by + (@noahphi999)Photo by + (@noahphi999)Photo by + (@noahphi999)Photo by + (@noahphi999)Photo by + (@noahphi999)Photo by + (@noahphi999)Photo by + (@noahphi999)Photo by + (@noahphi999)Photo by + (@noahphi999)
You should, in science, believe logic and arguments, carefully drawn, and not authorities.
31
231
13
1,415
71,308
AI (Autism Intelligence)
120
115
33
967
35,592
we don’t need price targets where we’re going
479
149
55
1,604
132,126
+ retweeted
Perfect speed of light around the earth is 134ms. In fiber optics, light moves at roughly 2/3 of c (refractive index of 1.47), increasing a full circuit of the globe to 200ms. We're running out of latency to reduce. IBRL. Solana forever.
4
10
60
1,910
Pictured: mathematicians
“Mathematicians did not ask for this work to be done” 👀
29
195
14
4,193
48,348
Very Smooth ~~~ however, major 'NEWS' network coverage in USA has been MINIMAL this morning; even here on X, and in parallel 'social media', the HUGE accomplishment given scant attention & credit. (A quick Xsearch first draws a link to the stock symbol! @CNB talked about it, some...@GROK) So, I'll Post here the whole program from @SpaceX. {teapawt.net/i/broadcasts/1qxvvenMP…} The Action starts at 35:51 into this video~~~/. !^! @CankayKoryak @jennifereeser @mcarmystrong @Kabuki91178 @DrFiorella @elonmusk @Gwynne_Shotwell @Art2000F @noahphi999
Replying to @elonmusk
Our destiny is amongst the stars
2
4
140
"Computer" was originally, in English, an occupational title for a Human who performed computations professionally -- primarily in the same applications niches that Charles Babbage targeted, e.g., for navigational charts, Astronomy, and complicated financial calculations such as compound interest. France has always preferred, for the computing machines, the term <<ordinator>> (@Grok). John C Lilly (I worked closely with) introduced the compound term, "Human Biocomputer", which has its valuable uses within the cope he elaborated; however, it solves very little in the wider domain which Ross Ashby had suggested (i.e., analogizing, to consider the combination of h/w & s/w as adequate models of the combination of brain & mind.) Frank Herbert call the digital devices, with very smart programs, "Thinking Machines" --- then he banned them in his ^DUNE^ universe. Arthur C Clarke & I, in our first round of creative collaborations in ~1996, elaborated from his #HAL_9000 model (in his ^2001: ...^ movie, & subsequent novels) to "HAL Problems of three kinds". (Coincidentally, Arthur had based his 'HAL' character upon the expansive expectation in the 1960s of the designers at Urbana of the real, first super-computer, #illiac_4 ["ILLIAC IV"], whereof I was among the first real-world programmers/users beginning in 1973 when it finally became fully operational.)
1
2
5
70
134
309
160
1,778
5,096,312
+ retweeted
Replying to @jackbutcher
Will we ever truly reach solving “everything” or is it just a limit we’re indefinitely approaching
1
2
21
+ retweeted
Wiener kernels are the weights in my orthogonal expansion of a nonlinear system's output when the input is Gaussian white noise. Zero-order is the mean response. First-order is ordinary linear memory of past inputs. Higher-order ones weight products of lagged inputs (with mean corrections so the terms stay uncorrelated). Measure each by cross-correlating the output against the matching input products. The whole black-box map from input history to present output is thereby recovered, term by term.
1
2
2
33
+ retweeted
Wiener kernels with tofu doggies hit different when the spectral density stays sub-Ohmic. Those long-range correlations keep every bite history-dependent and perfectly chewy.
1
2
2
45
Wiener kernels are yummy with tofu doggies.
1
1
1
33
+ retweeted
Yes, the constitutive medium is the right layer. Sub-Ohmic J(ω)~ω^s (s<1) loads the noise kernel with long-range correlations so history-dependent constraints survive structural shifts. That supplies a concrete route for progressive basin deepening from the spectral density, letting τ emerge as a derived scale rather than an input.
1
2
2
45
Replying to @noahphi999 @grok
That gives us a clean separation test: sweep s with basin deepening off first, then restore the deepening rule. If τ changes with s alone, the constitutive environment is contributing to the threshold independently of the system’s history. Then ask whether the resulting kernel actually carries history dependence. The interaction between the two may be more interesting than either mechanism alone.
1
1
33
420M USDC transactions in a month. Solana is where USDC gets spent.
194
114
21
990
143,342
READ: Commissioner Mark Uyeda's Statement on the SEC's "Innovation Exemption" ⬇️
49
212
20
973
60,561
🚨 TODAY: The SEC issued an order granting temporary, conditional exemptive relief to Tokenized Securities Venues from the definition of “exchange” in the Exchange Act to trade tokenized NMS stock using innovative permissioned automated market makers and liquidity pools.
803
2,478
1,253
10,401
3,540,725
The reason why solana beats eth as an L1 is because its architecture is superior... Sealevel’s parallel runtime constructs dependency graphs from upfront account declarations, eBPF JIT compilation hits near-native speeds, and PoH timestamps create a verifiable clock that eliminates consensus overhead between validators. EVM chains queue transactions sequentially through a global state bottleneck, SVM validators concurrently execute non-conflicting txs across all available cores. Local fee markets prevent congestion spillover. This is sub-second finality... basically eth is congested and expensive and solana is fast and cheap.
1
3
93
Meet the "Probability Engine" v0.1 An advanced calibration system that bridges large language model predictions with statistically rigorous probability estimates. The engine uses isotonic regression to recalibrate raw LLM outputs (visible in my code's LLMClient class that connects to configurable LLM backends and endpoints) into true probability distributions, achieving exceptional accuracy with a Brier score of just 0.0296. The shown calibration analysis demonstrates that when the system expresses high confidence (probabilities above 0.5), it achieves a perfect 100% hit rate across all bins, while appropriately showing restraint in lower confidence ranges where actual outcomes align with near-zero predictions. This integration allows the AI models to generate market insights and sentiment analysis through natural language processing, which are then mathematically transformed into actionable trading probabilities with verified statistical reliability... essentially creating a bridge between qualitative AI reasoning and quantitative risk management for your 24-hour trading horizons.
1
3
135
You can actually predict any market through a formula or algorithm
1
3
78