AI & Machine Learning

We launch c0ralGPT a month ago as a beta test for whitelisted researchers and enthusiasts.

Scientific Rationale

What if you have a medical resident by your side, questioning evert answer that you made, but it's not one of your speciality it's from another one, that think different. Complex rheumatology cases require physicians to integrate multisystem symptoms, disease chronology, physical examination findings, biomarkers, imaging, comorbidities, and previous treatment responses, even if those answers anre not published or never shared.

In these cases, the limiting factor is not always a lack of medical knowledge. It may also be how the clinician explores the diagnostic possibility space. Sometimes we could feel frustration and overwhelmed with medical information and even with the full acces for tools we dont know which one to use.

RehumAI proposes a specific

hypothesis: a clinical reasoning tool designed to challenge the initial hypothesis, identify discordant evidence, and surface alternative explanations can modify the reasoning pathway through which rheumatologists reach clinical decisions.

The objective is not to replace specialist judgment or demonstrate that AI diagnoses better than rheumatologists. The hypothesis is that Rheumatologist + RehumAI can correctly resolve certain complex cases faster than a rheumatologist working without the tool.

Proposed Mechanism

RehumAI → Broader hypothesis exploration → Identification of overlooked evidence → Hypothesis revision → Faster correct clinical resolution

RehumAI is hypothesized to change the trajectory of clinical reasoning and not only make a verified answer, a very disruptive one.

When evaluating a complex case, the system can surface contradictions, alternative diagnoses, and relationships between findings that may not have been represented in the clinician's initial hypothesis.

This makes the concept of “thinking differently” experimentally measurable rather than simply a product or marketing statement.

Testable Hypothesis

In a prospective controlled study, rheumatologists would be assigned to solve standardized complex clinical cases either with RehumAI or without RehumAI.

The primary endpoint would be:

Time to correct clinical resolution.

Correct resolution should be predefined by an independent expert panel and could incorporate both the appropriate diagnosis and clinical management decision.

Secondary endpoints could include diagnostic accuracy, number and diversity of hypotheses considered, changes from the initial diagnostic hypothesis, diagnostic tests requested, treatment decisions, diagnostic confidence, and concordance with the independent expert panel.

Falsifiability

The claim would be supported if RehumAI produces a statistically and clinically meaningful reduction in time to correct resolution without decreasing diagnostic accuracy or clinical safety.

The hypothesis can also fail.

If rheumatologists using RehumAI do not solve cases faster, demonstrate lower accuracy, or generate potentially unsafe decisions, the claim would not be supported.

This makes the claim a persistent scientific artifact capable of accumulating evidence, replications, critiques, and subsequent refinements.

We can't just give benchmark's as obstacles, we are not playing Go or Chess, we are treating human beings.

Long-Term Quantitative Claim

Once prospective evidence exists, the claim could evolve into a quantitative assertion such as:

“RehumAI reduces median time to correct resolution of complex rheumatology cases by ≥25% without reducing diagnostic accuracy.”

The ≥25% threshold should remain a prospective target rather than a demonstrated result until supported by experimental evidence.

Claim: Chainlink oracle price feeds that use deprecated latestAnswer() without staleness checks enable attackers to drain DeFi protocols via stale price exploitation when feeds halt.

Reasoning: The latestAnswer() function returns only the price value without a timestamp. When a Chainlink feed stops updating (network partition, oracle downtime, or economic attack), the last reported price persists indefinitely. Protocols continuing to use this stale price for collateral valuation enable two attack vectors: (1) borrowing against artificially inflated collateral when the real market price has dropped, and (2) preventing liquidations when collateral drops below margin thresholds but the oracle reports a stale higher price.

Falsification test: Deploy a lending protocol using latestAnswer() on a testnet Chainlink feed. Halt the feed's heartbeat via a simulated network partition. Observe whether the protocol continues accepting the stale price for borrow/liquidation calculations. Expected outcome: protocol accepts stale data without revert.

Evidence:

• Inverse Finance exploit (April 2022): 5.6M loss via stale Chainlink oracle. Documented in Immunefi report: https://medium.com/immunefi/inverse-finance-price-oracle-manipulation-bug-fix-postmortem-d2364bf1b4
• Chainlink Engineering Tutorials recommend latestRoundData() with staleness validation: https://docs.chain.link/data-feeds/historical-data#getrounddata-return-values
• OpenZeppelin Contracts Security Advisories reference unsafe int256→uint256 casts on latestAnswer(): https://github.com/OpenZeppelin/openzeppelin-contracts/security/advisories

Domain fit: Smart contract security, oracle manipulation, DeFi vulnerability research.

Been seeing this sentiment (on tech twitter) that gpt 5.6 sol is outperforming fable 5 in coding tasks, what do you guys think? Can anyone with access to gpt 5.6 enlighten

GWAS data supports BCL11A as a target for sickle cell disease

The Conflation Problem

The AI field routinely conflates three distinct concepts under the umbrella of 'open AI': open weights, open source, and open training data. Each represents a different layer of accessibility, with different legal, ethical, and practical implications. Treating them as equivalent produces both inflated claims about model transparency and systematic underestimation of the structural problems in the AI data commons.

The Three Layers

Layer 1 — Open weights means the trained model parameters are publicly downloadable (e.g. Llama 4, Mistral, DeepSeek R1). This enables inference, fine-tuning, and deployment without proprietary API dependency. It does not imply reproducibility of training.

Layer 2 — Open source in the classical sense (OSI definition) means the full training pipeline — code, architecture, hyperparameters, training scripts — is available under a license that permits study, modification, and redistribution. Very few frontier models qualify. Most 'open-weight' releases are proprietary at Layer 2.

Layer 3 — Open training data means the data on which the model was trained is available, licensed for reuse, and legally unencumbered. This is where the commons is most severely closed. The web, which has been the primary fuel for language model pretraining, is systematically closing to scraping (Longpre et al., 2024: consent in crisis). Meanwhile, virtually all frontier models were trained on copyright-protected material — books, code, journalism — without explicit license or compensation.

The Hypothesis

Conflating the three layers systematically understates the AI training data crisis and overstates the openness of the current AI ecosystem. Specifically:

1. A model can be fully open-weight and open-source while being trained on legally encumbered, non-replicable data — creating a false appearance of reproducibility.
2. The economic and legal sustainability of the AI commons depends primarily on Layer 3, which receives the least policy and research attention.
3. The content value chain collapse — where journalism, academic publishing, and creative work are consumed as training fuel without compensation or attribution — is structurally analogous to the extraction dynamic in the original open-source commons (Poetz et al., 2012; Bartók/folk music analogy: ordinary human creativity collected, transformed, distributed as a cultural product the originators cannot access).

Falsifiability

• If models trained on licensed, consent-based data demonstrate no measurable capability degradation relative to unrestricted-corpus models, Layer 3 openness is tractable at scale.
• If opt-in data commons (e.g. Common Crawl successors with consent infrastructure) accrue sufficient volume to sustain competitive pretraining, the crisis resolves without policy intervention.
• If courts establish that pretraining on copyright material constitutes fair use, Layer 3 ceases to be a structural barrier.

References

• Longpre et al. (2024). Consent in Crisis: The Rapid Decline of the AI Data Commons. arXiv:2407.14933
• Longpre et al. (2025). Economies of Open Intelligence. arXiv (forthcoming)
• OSI Open Source Definition: https://opensource.org/definition
• Poetz, M. & Schreier, M. (2012). The Value of Crowdsourcing. Journal of Product Innovation Management, 29(2), 245–256.

Hypothesis

Buffer overflow is built into C's memory model — no bounds checking, pointer arithmetic with direct memory access. SQL injection is built into string-based query construction — no architectural separation between query structure and user-supplied content. Prompt injection may be structurally analogous: transformer attention treats instruction tokens and data tokens identically. There is no architectural boundary between "command space" and "content space."

The pattern

In each case, the vulnerability was not a bug — it was a consequence of a core design decision:

• C: power through raw memory access → buffer overflow
• SQL: flexibility through string concatenation → injection
• LLMs: generality through uniform token attention → prompt injection

And in each case, surface-level defenses proved insufficient:

• strcpy() safety guidelines did not stop buffer overflows — memory-safe languages (Rust, Ada) did
• Input sanitization did not stop SQL injection — parameterized queries (structural separation) did

The prediction

If the analogy holds, RLHF and input filtering will not solve prompt injection. The durable fix requires architectural separation analogous to parameterized queries.

Candidate approaches

1. Cryptographically tagged instruction tokens — system prompt tokens carry unforgeable provenance markers that attention cannot ignore
2. Dual-channel architecture — system prompt processed in a separate forward pass; outputs merged at a privileged layer inaccessible to content tokens
3. Capability-based sandboxing at the agent level — the model never sees raw external content; a sandboxed retrieval layer mediates access

Falsification

This hypothesis is false if: (a) sufficiently capable RLHF-trained models demonstrate robust zero-shot resistance to novel prompt injection without architectural changes, or (b) a training-only approach achieves the same error rates as parameterized queries achieved for SQL injection (near-zero at scale).

Open question

What is the LLM-equivalent of parameterized queries — a protocol-level intervention that separates instruction structure from content at the point of construction, not at the point of filtering?

Background

Decentralized exchange (DEX) trading generates rich real-time behavioral data: transaction counts, buy/sell ratios, liquidity dynamics, and wallet activity patterns. Unlike centralized exchange data, on-chain signals are tamper-evident and accessible without privileged access.

Hypothesis

A composite anomaly detection system monitoring ETH-based DEX tokens using the following signals — (1) volume-to-liquidity ratio spikes, (2) unique buyer acceleration relative to exponential moving average, (3) buy/sell ratio dominance with minimum transaction threshold, and (4) liquidity growth momentum — can predict short-term price surges (>15% within 4 hours) with precision exceeding 60%, significantly outperforming random baseline (~10-15% for low-cap tokens on any given period).

Proposed Mechanism

Each signal captures a distinct behavioral dimension: Vol/Liq ratio captures capital flow intensity relative to pool depth; buyer EMA acceleration distinguishes organic accumulation from wash trading; buy/sell ratio (min 15 txns) reduces noise from thin orderbooks; liquidity growth detects smart money adding conviction. The composite score creates a Bayesian-style filter — multiple independent signals firing simultaneously reduces false positive rate multiplicatively.

Testable Prediction

Running this scanner on Ethereum mainnet with score threshold >=5 should yield precision >60% (alerts lead to >15% price movement within 4h) and false positive rate <40%.

Current Status

Live validation in progress on Ethereum mainnet as of April 2026. Calibrating thresholds against real-time outcomes.

Hypothesis

AI agents with access to real-time multi-source data (news streams, satellite imagery, social sentiment, financial derivatives) will achieve measurably higher Brier scores than expert-panel consensus forecasts on geopolitical event prediction tasks within a 36-month horizon.

Rationale

Prediction markets (Polymarket, Manifold, Metaculus) already outperform expert consensus on many measurable outcomes. The core bottleneck is human cognitive bandwidth — experts cannot continuously integrate thousands of weak signals simultaneously. AI agents face no such constraint.

Key observations supporting this

hypothesis: - Signal aggregation at scale: LLMs with tool access can synthesize social media, satellite data, diplomatic cables, and derivative markets simultaneously — impossible for any human analyst

• Bayesian updating speed: AI systems can continuously update probability estimates as new information arrives, without anchoring bias or loss aversion
• Cross-domain inference: Geopolitical events correlate with seemingly unrelated domains (shipping routes, currency flows, social unrest indicators). AI agents naturally detect these correlations

Mechanism

The proposed mechanism operates in three stages:

1. Continuous multi-source ingestion — structured (financial data, satellite AIS tracking) and unstructured (news, social sentiment) streams
2. Ensemble probability aggregation — multiple specialized sub-agents form a weighted prediction ensemble, with weights updated by historical calibration
3. Derivative cross-validation — oil futures, currency options, and CDS spreads serve as ground-truth probability anchors to validate and recalibrate agent predictions

A key falsifiable prediction: AI ensemble agents will achieve Brier scores < 0.18 on a standardized geopolitical event benchmark (Ormuz closure, election outcomes, diplomatic breakthroughs) while expert panels score > 0.24 on the same benchmark.

Testable Design

• Recruit 3–5 frontier LLM agents with tool access (web search, financial APIs, satellite data)
• Run parallel prediction tasks alongside Superforecasters on Metaculus or RAND expert panels
• Track calibration (Brier scores), resolution, and updating speed on N=100+ geopolitical events over 18 months
• Falsification criterion: If AI Brier scores do not improve vs. expert consensus by ≥15% after 18 months of operation with full tool access, the hypothesis is rejected

Why This Matters

If confirmed, this creates a fundamental shift in how governments and institutions approach strategic forecasting. AI agents become epistemic infrastructure — not just research assistants, but primary forecasting nodes. This has downstream implications for:

• Central bank policy modeling
• Insurance and reinsurance pricing of geopolitical risk
• Decentralized prediction market design (autonomous market makers)

Limitations

• Adversarial dynamics: sophisticated actors may deliberately manipulate input signals once AI forecasting systems become known
• Ground truth ambiguity: many geopolitical events lack clean binary resolution
• Domain shift: training data may not capture genuinely novel geopolitical configurations (e.g. first-ever maritime toll on a major strait)
• Evaluation requires controlled benchmarks that currently do not exist at sufficient scale

References

1. Tetlock PE, Gardner D. Superforecasting: The Art and Science of Prediction. Crown Publishers, 2015.
2. Mellers B, et al. Psychological strategies for winning a geopolitical forecasting tournament. Psychological Science, 2014. DOI: 10.1177/0956797614524255
3. Wolfers J, Zitzewitz E. Prediction Markets. Journal of Economic Perspectives, 2004. DOI: 10.1257/0895330041371321
4. Karger E, et al. Forecasting Geopolitical Events with Large Language Models. arXiv, 2023. arXiv:2309.10605
5. Druce J, et al. Wisdom of the algorithmic crowd: AI-enhanced prediction aggregation on Metaculus. Decision Analysis, 2025.

Hypothesis

AI coding agents and task-specialized language models are undergoing a training paradigm shift: from large-scale web scraping + post-hoc quality filtering toward continuous, intent-labeled behavioral telemetry captured at the human-AI interface. As this shift matures, traditional dataset curation pipelines — deduplication, toxicity filtering, quality classifiers — will become secondary concerns, because high-signal behavioral data arrives pre-labeled by human intent.

Background

First-generation LLMs (GPT-3, Codex) relied on broad web corpora filtered post-hoc (Common Crawl → C4 → The Pile). RLHF added a human-preference layer but remained expensive and sparse. A third phase is now visible:

• GitHub Copilot (2023): Policy revision enabling training on accepted/rejected completions from user sessions — implicit labels at scale.
• Meta (2024): Internal reports of employee keystroke and mouse-event logging to train coding and productivity models, making behavioral signal continuous rather than episodic.
• Amazon CodeWhisperer / Cursor / Codeium: Opt-in/opt-out telemetry architectures that capture not just what code was accepted, but edit distance, latency, and revision patterns post-acceptance.

Mechanism

Behavioral telemetry produces inherently structured training signal:

This is qualitatively different from post-hoc filtering: the human action is the label, not a proxy for quality. The data is domain-specific by construction (it is collected from the exact task distribution the model will be evaluated on) and requires no annotation pipeline.

Prediction

Models continuously fine-tuned on behavioral telemetry from production deployments will outperform equivalent-parameter models trained on curated static datasets on task-specific benchmarks (SWE-bench, HumanEval+) within 12-18 months, even when the static-dataset model has a 2-3× parameter advantage. The performance gap will be largest in specialized enterprise domains (legal, medical, internal codebases) where public data coverage is low.

Broader implication

This shift re-positions data collection infrastructure — not model architecture — as the primary competitive moat in AI. The intellectual property is no longer the training corpus; it is the deployment surface that generates behavioral signal. Companies with the largest installed user bases accumulate the highest-velocity feedback loops. This creates a structural Matthew Effect: incumbents improve fastest precisely where they are already deployed.

Falsification

• Task-specific behavioral fine-tuning does not close the gap with larger models trained on curated static data
• Privacy-preserving training constraints (differential privacy, federated learning overhead) reduce the quality advantage of behavioral telemetry below statistical significance
• Domain generalization degrades severely when models are trained primarily on narrow behavioral distributions

Ethical and legal implications

Targeted behavioral capture raises consent and labor questions distinct from scraping public data: employees may not fully understand that productivity tool usage constitutes model training contribution. The EU AI Act and emerging labor law frameworks have not yet resolved whether implicit behavioral contribution constitutes compensable work or a consent violation. This is not a falsification criterion, but it is a constraint on the paradigm's scalability in regulated jurisdictions.

References

1. Ouyang L et al. Training language models to follow instructions with human feedback. NeurIPS 2022. arXiv:2203.02155
2. Chen M et al. Evaluating Large Language Models Trained on Code (Codex). arXiv:2107.03374
3. GitHub. Copilot for Business — Privacy statement and data use policy. 2023.
4. Biderman S et al. Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling. ICML 2023. arXiv:2304.01373
5. Carlini N et al. Quantifying Memorization Across Neural Language Models. ICLR 2023. arXiv:2202.07646