Jason Calacanis
B70/ 100
3 claims reviewed
Corrects the school-policy description; the productivity claim rests on his own experience.
Evidence balance 8/10What the school policy actually says
The 30-second take
Jason correctly narrows the description of New York’s AI restriction. The panel provides much less support for its claims about AGI, lasting learning gains and politicians’ motives.
Episode: GPT-6 Hits AGI? Tech Euphoria 2.0, SF Mansion Shortage, NYC Bans AI in Schools & Venezuela Oil Deal
Speaker assessments
David Sacks
F45/ 100
4 claims reviewed
Access concerns are reasonable. Motive claims are unsupported.
Evidence balance 3/10David Friedberg
D50/ 100
4 claims reviewed
Cites the education review accurately but draws conclusions it cannot establish.
Evidence balance 3/10Chamath Palihapitiya
F45/ 100
3 claims reviewed
Proposes adaptive teaching but supplies no test for the AGI or learning-style claims.
Evidence balance 4/10Scores, sample sizes & reasoning
| Speaker | Overall | Accuracy 30% | Coherence 20% | Evidence support 20% | Calibration 15% | Evidence balance 15% | Claims |
|---|---|---|---|---|---|---|---|
| Jason Calacanis | 70 / B68 before rounding | 7 | 7 | 6 | 6 | 8 | 3 |
| David Sacks | 45 / F45.5 before rounding | 5 | 6 | 4 | 4 | 3 | 4 |
| David Friedberg | 50 / D47.5 before rounding | 5 | 6 | 5 | 4 | 3 | 4 |
| Chamath Palihapitiya | 45 / F47 before rounding | 5 | 6 | 4 | 4 | 4 | 3 |
Jason Calacanis
Credit. Corrects the scope of the school restriction.
Deduction. Personal workflow gains and financing advice lack comparative evidence.
Evidence balance 8/10
Corrects an exaggerated policy description and adds stage-specific financing qualifications.
Credit. Distinguishes the younger-grade moratorium from the high-school pilots. Scope of the school policy
Credit. Refines financing advice after the discussion introduces company-stage differences. Raising capital during a window
No concrete selective presentation was found in the reviewed sample.
David Sacks
Credit. Recognizes stage-dependent financing risks and unequal technology access.
Deduction. Infers harmful political intent and model-market dominance too readily.
Evidence balance 3/10
The political interpretation gives little weight to the policy’s stated educational rationale.
Credit. Distinguishes early-stage and mature-company liquidity decisions. Founder liquidity
Deduction. The policy gives developmental and instructional reasons for limits and includes supervised high-school pilots. Those details provide a competing explanation that must be addressed before inferring deliberate dependence or harm. Political motives for a school restriction
David Friedberg
Credit. Cites a real education evidence review.
Deduction. Overstates the historical contrast and infers safety and political causes from thin evidence.
Evidence balance 3/10
Selects the encouraging side of mixed research and makes an overly clean historical contrast.
Credit. Acknowledges that some assisted gains fade when the tool is removed. AI in education evidence
Deduction. The cited review emphasizes short-term evidence and gaps in US K–12 causal research. These limits undermine a broad inference about the absence of educational harm. Absence of evidence of harm
Deduction. Amazon reported about $1.64B in sales in 1999. Real revenue cannot by itself distinguish the current market from the dot-com period. Dot-com revenue contrast
Chamath Palihapitiya
Credit. Emphasizes adaptation to student needs.
Deduction. AGI, stable cyber equilibrium and learning-style matching lack sufficient validation.
Evidence balance 4/10
The education thesis relies on learning-style matching without confronting contrary research.
Credit. Recognizes that students differ in pace and instructional needs. Personalized learning styles
Deduction. The learning-styles review found inadequate evidence that matching visual or auditory preferences improves learning. Personalized pace and feedback may help, but they do not validate that specific matching premise. Personalized learning styles
| Speaker | Facts | Opinions / causal arguments | Forecasts |
|---|---|---|---|
| Jason Calacanis | 1 | 2 | 0 |
| David Sacks | 0 | 3 | 1 |
| David Friedberg | 2 | 2 | 0 |
| Chamath Palihapitiya | 0 | 2 | 1 |
Claim reviews
14 claims · 4 speakers
01AGI has arrivedDefinition-dependent
The discussion leaves the meaning of AGI undefined.
The claim paraphrased
Artificial general intelligence is already here.
Evidence in the episode
The claim has no agreed operational threshold in the discussion.
Why this judgment
The discussion never specifies which tasks, level of autonomy or reliability would count as AGI. Without those requirements, another evaluator cannot reproduce the judgment.
Confidence in assessment~85%
The episode supplies no operational definition against which to check the claim.
What would change this judgment?
A stated definition, test suite and independently reproduced results covering its requirements.
02Dot-com revenue contrastToo absolute
Amazon had substantial revenue during the dot-com boom.
The claim paraphrased
The dot-com era relied on non-dollar metrics, unlike today’s real AI revenues.
Evidence checked
Amazon reported roughly $1.64 billion in 1999 net sales.
Why this judgment
Real revenue existed during the dot-com period, so its presence cannot by itself distinguish today from a speculative cycle. A useful comparison would examine growth, margins, capital intensity and valuation.
Confidence in assessment~95%
Amazon’s 1999 financial statements directly contradict the broad historical contrast.
Evidence-balance issue
Amazon reported about $1.64B in sales in 1999. Real revenue cannot by itself distinguish the current market from the dot-com period.
What would change this judgment?
A like-for-like historical sample and valuation analysis.
03AI in education evidenceSupported with limits
Some studies found that gains faded when AI assistance was removed.
The claim paraphrased
A Stanford review found that assisted learning gains can fade when AI assistance is removed.
Evidence checked
Stanford’s review identifies 20 causal studies and reports that some gains fade without assistance.
Why this judgment
The review supports Friedberg’s description of gains fading. It also reports limited long-term evidence and no causal US K–12 studies in its sample, which restricts what it can tell us about schools generally.
Confidence in assessment~95%
The cited review reports this result while explicitly identifying the gaps in its evidence base.
What would change this judgment?
Long-term causal studies in representative US K-12 settings.
04Scope of the school policySupported
The policy is narrower than a blanket school AI ban.
The claim paraphrased
New York’s moratorium targets younger students and permits limited high-school AI pilots.
Evidence checked
The September 2 announcement applies the student-facing generative-AI moratorium to grades 2-K–8 and allows limited high-school pilots for up to 50,000 students.
Why this judgment
Jason’s correction improves the discussion by separating age groups and use cases. His K–8 shorthand omits the younger 2-K coverage, but the central distinction from an across-the-board ban holds.
Confidence in assessment~95%
Very high confidence in the dated city announcement.
What would change this judgment?
A revised policy changing the age coverage or permitted pilots.
05Political motives for a school restrictionWeak support
A policy disagreement does not establish that motive.
The claim paraphrased
Political leaders may benefit from keeping students dependent and downwardly mobile.
Evidence checked
The episode infers possible electoral benefit from the restriction. The city’s stated reasons concern development, safety and instruction; neither statement establishes private intent.
Why this judgment
The policy could be mistaken without being designed to harm students. Inferring intent requires evidence that distinguishes that explanation from sincere caution, coalition pressure or an incorrect assessment of the research.
Confidence in assessment~95%
Very high confidence that the evidence presented does not establish the suggested motive.
Evidence-balance issue
The policy gives developmental and instructional reasons for limits and includes supervised high-school pilots. Those details provide a competing explanation that must be addressed before inferring deliberate dependence or harm.
What would change this judgment?
Contemporaneous communications or decisions that distinguish deliberate harm from the alternatives.
06Personalized learning stylesMixed
Matching visual or auditory preferences has not shown the claimed learning benefit.
The claim paraphrased
AI instruction tailored to visual or auditory learning styles should produce better education.
Evidence checked
The Pashler and colleagues review found inadequate support for matching instruction to claimed learning styles. This is separate from adapting difficulty, pace or feedback.
Why this judgment
Students differ, but preferences do not establish which presentation improves learning. The case for adaptive tutoring is stronger when based on prior knowledge and measured progress than fixed visual or auditory labels.
Confidence in assessment~90%
High confidence in the distinction between preferences and demonstrated instructional benefits.
Evidence-balance issue
The learning-styles review found inadequate evidence that matching visual or auditory preferences improves learning. Personalized pace and feedback may help, but they do not validate that specific matching premise.
What would change this judgment?
Randomized evidence that the proposed matching improves retained learning beyond simpler adaptive methods.
07Cybersecurity equilibriumPlausible, unresolved
The forecast leaves stability and its timing undefined.
The claim paraphrased
AI-enabled attackers and defenders may converge on a stable equilibrium.
Evidence in the episode
Adaptive competition can produce an equilibrium, an arms race or intermittent disruption.
Why this judgment
The episode gives no reason to favor a stable equilibrium over an arms race or repeated disruption. A testable forecast would specify a time horizon and what counts as stability.
Confidence in assessment~85%
No model or measured trend in the discussion distinguishes equilibrium from the other outcomes.
What would change this judgment?
Measured defensive and offensive trends and a falsifiable equilibrium threshold.
08Frontier duopolyInsufficiently demonstrated
The discussion does not establish a lasting two-provider market.
The claim paraphrased
OpenAI and Anthropic lead the frontier while other models are becoming commodities.
Evidence in the episode
A durable duopoly requires evidence across relevant tasks, prices, distribution and switching costs.
Why this judgment
The conversation does not provide a consistent comparison covering those dimensions. A lead on selected tasks would not prove that every other provider lacks differentiation.
Confidence in assessment~85%
The panel supplies no consistent comparison of providers across tasks, prices and customer switching costs.
What would change this judgment?
Independent task-specific evaluations and market evidence over a defined period.
09Founder liquidityWell calibrated
A founder’s liquidity decision depends on the company’s stage.
The claim paraphrased
Secondary sales should be evaluated differently at early and mature startup stages.
Evidence in the episode
The stage distinction matters: financing capacity, concentration and remaining execution risk differ.
Why this judgment
Company stage changes the risks of a secondary sale. The discussion identifies relevant considerations but does not establish an optimal amount for a particular founder.
Confidence in assessment~85%
Sacks qualifies the advice by company stage, financing capacity and execution risk.
What would change this judgment?
Company-specific financing terms and incentives could change the recommendation.
10Personal productivity liftUnverified anecdote
Jason’s reported improvement has not been independently measured.
The claim paraphrased
An AI assistant improved his own workflow materially.
Evidence in the episode
Jason describes improvements in his own workflow without providing a task benchmark.
Why this judgment
Jason gives no baseline, timing measurements or repeated comparison. His account could justify testing the assistant on similar tasks, but it cannot establish how much other users would gain.
Confidence in assessment~90%
The episode offers only a self-report, with no independent measurement of time or work quality.
What would change this judgment?
A documented before-and-after task benchmark with quality and time measured.
11Raising capital during a windowConditional advice
The decision depends on runway, dilution and the founder’s incentives.
The claim paraphrased
Founders should consider raising available capital and taking some liquidity as opportunities permit.
Evidence in the episode
Financing can extend runway and reduce personal concentration, while dilution and incentives carry costs.
Why this judgment
The discussion’s stage-related qualifications improve the argument. No universal financing rule follows from a favorable market window.
Confidence in assessment~85%
The discussion acknowledges stage differences but gives no company-specific calculation of runway or dilution.
What would change this judgment?
A company-specific runway, dilution and execution-risk analysis.
12AI access and the digital dividePlausible, unproved
The predicted education gap depends on whether these uses of AI improve learning.
The claim paraphrased
Restricting public-school AI while private schools use it will widen the education gap.
Evidence checked
The city policy limits student-facing use in younger grades. The episode does not compare actual private-school adoption, instructional quality or learning outcomes.
Why this judgment
An effective tool available only to some students could widen a gap. That requires the tool to improve learning in the relevant setting; access alone does not establish the size or direction of the outcome.
Confidence in assessment~80%
Moderate-to-high confidence in the conditional risk; the projected achievement effect is not established.
What would change this judgment?
Comparable learning outcomes across schools, adjusting for resources, prior achievement and how AI is used.
13Absence of evidence of harmToo broad
A thin evidence base cannot settle safety or effectiveness.
The claim paraphrased
The education literature provides little basis for thinking AI harms children’s learning.
Evidence checked
The cited Stanford review stresses short-term studies and important research gaps, including US K–12 causal evidence. It also distinguishes performance with a tool from learning that persists without it.
Why this judgment
The limited research cannot rule out harm across ages and uses. Studies need to measure which tools help which students, including whether gains persist when the assistance is removed.
Confidence in assessment~90%
High confidence that the broad inference exceeds the review’s scope.
Evidence-balance issue
The cited review emphasizes short-term evidence and gaps in US K–12 causal research. These limits undermine a broad inference about the absence of educational harm.
What would change this judgment?
Long-term studies measuring independent learning and development across defined uses.
14Teachers’ union explanationWeak support
The discussion supplies no record linking union demands to this policy.
The claim paraphrased
Teacher unions’ fear of automation is a major reason for the school AI restriction.
Evidence in the episode
Friedberg presents an incentive-based explanation without documentary evidence linking union demands to the specific restriction.
Why this judgment
A group can have an economic interest without that interest causing a particular policy. Parent preferences, pedagogical uncertainty and privacy concerns are competing explanations that need to be tested.
Confidence in assessment~90%
High confidence in the missing causal evidence; no finding is made about private motives.
What would change this judgment?
Negotiating records, policy drafts or attributable decision-maker accounts showing the claimed influence.
No claims match these filters.
Source desk7 references
- All-In / LibsynOfficial episode
Original episode, published 2026-09-04.
- All-In / YouTubeWatch the episode
Original recording; timestamps mark discussion chapters.
- 1% BetterAutomated transcript
Automated transcript. Attribution is provisional; links mark discussion chapters.
Review of the evidence for AI use in schools.
Historical financial statements used to check the Amazon comparison.
- New York City Mayor’s OfficeNYC student-facing AI policy
September 2, 2026: 2-K through grade 8 moratorium, with limited high-school pilots.
- Pashler, McDaniel, Rohrer & BjorkLearning Styles: Concepts and Evidence
Research review distinguishes presentation preferences from demonstrated learning benefits.
How the scores work
Five dimensions, one weighted score
Each dimension is scored from 0–10. The weighted total is multiplied by ten and rounded to the nearest five, with halves rounded up.
Grades use that rounded score: A ≥80, B ≥70, C ≥60, D ≥50, F <50.
Accuracy tests factual correctness; coherence tests whether the conclusion follows; evidence support tests the strength of the sources. Calibration asks whether the speaker’s certainty matches the evidence.
Accuracy30%
Coherence20%
Evidence support20%
Calibration15%
Evidence balance15%
Evidence balance
Balance assesses cherry-picking, omitted counterevidence, inconsistent standards and distorted alternatives. Higher scores mean fairer treatment of evidence. Each deduction names the affected claim, the missing evidence and why it changes the conclusion.
Political disagreement and presumed intent do not count. A false claim or weak source alone does not establish cherry-picking.
- 9–10Actively tests strong contrary evidence and represents it fairly.
- 7–8Generally fair selection and relevant qualifications; no material distortion identified.
- 5–6Mixed: fair treatment in some claims, material omissions in others.
- 3–4Materially selective samples, comparisons or treatment of contrary evidence.
- 0–2Repeated, severe distortion or dismissal of directly relevant counterevidence.
What these scores can tell you
Scores assess selected arguments. They do not measure a speaker’s overall reliability or percentage of correct statements. Confidence ratings are subjective; no statistical intervals are calculated. Automated-transcript attribution remains provisional. Small differences are not statistically meaningful, and no independent second rater has checked the scores.
New reviews select up to 20 significant claims, without speaker quotas. The sample is editorial, not random or exhaustive. All-In #286–292 retain their original 14-claim samples. An absent or unassessable speaker has no score and leaves a gap in the chart.
Primary sources support the assessments. Current product and methodology pages are labeled when used retrospectively. Unresolved forecasts are assessed for their support. All-In video links use chapter starts; Pivot links use transcript timestamps, which can differ from ad-supported audio.
This review: 14 selected claims. Published Sep 4, 2026; reviewed Oct 5, 2026.
Rubric v2.0. Earlier scores and correction records remain in the review data.