ClaimsPeople

Balaji Srinivasan

Balaji S. Srinivasan is a technology entrepreneur, investor and author who was formerly CTO of Coinbase and a general partner at Andreessen Horowitz, and who founded Network School.

1,569 claims collected, 251 researched, 80 resolved. Report of 4 November 2026, method 0.1.0+c876030c.

Scores by kind of claim

Each kind of claim is scored on its own, out of 100: the share that held up, with the range it likely falls in. Fewer resolved claims give a wider range.

  • Factual claims

    Statements about the present or the past, checked against the record: did they hold up?

    72 65 79

    High confidence, 60 factual claims resolved. From 15 documents, 2018 to 2025; 19 documents sampled from 34 candidates found.

    Not scored: 35 unresolvable, 2 checkable only against their own records.

  • Predictions

    Statements about events outside the person's control, checked once they fell due: did they come true?

    66 54 78

    Medium confidence, 19 predictions resolved. From 5 documents, 2018 to 2024; 19 documents sampled from 34 candidates found.

    Not scored: 111 unresolvable. 21 more fall due later.

    Does not yet allow for how likely each prediction was when made.

  • Promises

    Commitments to the person's or organisation's own future action, with something to deliver: was it delivered?

    Insufficient record: 1 promise resolved, 5 needed for a score.

    Not scored: 2 unresolvable.

All types combined: 71 (likely 64 to 77, high confidence), from 79 claims: 60 factual claims, 19 predictions. It mixes kinds of claim in different proportions for different people, so compare people kind by kind.

Balaji Srinivasan's claims

60 claims, sorted by newest said. Choose a claim to read it in full.
VerdictKindSaidClaimTopic
Mostly false Factual 22 Dec 2025 Electric vehicles now fill up roughly as fast as gasoline vehicles. Factual · 22 Dec 2025 Technology
Mostly true Factual 22 Dec 2025 Gold can be stopped at TSA metal detectors. Factual · 22 Dec 2025 Other
Resolved true Factual 12 Dec 2025 It has been reported that the Western left's shift toward abstract expressionism and deconstructionism was partially sponsored by the CIA. Factual · 12 Dec 2025 Philosophy, ethics and society
Resolved true Factual 4 Dec 2025 Tech unicorn founders are much rarer than professional athletes. Factual · 4 Dec 2025 AI, the economy and work
Mostly true Factual 12 Oct 2025 Olaf Carlson-Wee sees network states speed-running the path from bare land to Dubai and Singapore. Factual · 12 Oct 2025 Their own work or organisation
Resolved true Factual 12 Oct 2025 Jacqueline Poh runs JTC, which is Singapore's $40B real estate arm. Factual · 12 Oct 2025 Economics and markets
Mostly false Factual 12 Oct 2025 The Solana network state recently negotiated with governments for a measure of diplomatic recognition. Factual · 12 Oct 2025 Other
Mostly true Factual 12 Oct 2025 In 2024 Balaji Srinivasan held a conversation with Bryan Johnson comparing their paths, which led to similar conclusions. Factual · 12 Oct 2025 Their own work or organisation
Mostly true Factual 3 Aug 2025 Killer AI already exists in the form of drones, and every country is pursuing it. Factual · 3 Aug 2025 AI policy and governance
Mostly true Factual 3 Aug 2025 AI can break captchas but cannot fake onchain balances. Factual · 3 Aug 2025 AI capabilities
Mostly true Factual 3 Aug 2025 Current AI agents cannot set complex goals or properly verify outputs. Factual · 3 Aug 2025 AI capabilities
Mixed Factual 3 Aug 2025 AI is currently having a decentralizing rather than centralizing effect on the industry. Factual · 3 Aug 2025 AI, the economy and work
Mostly false Factual 3 Aug 2025 AI provably cannot solve chaotic, turbulent, or cryptographic equations. Factual · 3 Aug 2025 AI capabilities
Mostly true Factual 28 Jul 2025 Bitcoin is valued everywhere there is an internet connection. Factual · 28 Jul 2025 Economics and markets
Mixed Factual 28 Jul 2025 The keys are the keys to the car itself: a digital signature starts the engine. Factual · 28 Jul 2025 Technology
Mostly true Factual 28 Jul 2025 Doors, including house doors, can be secured with an onchain smart lock controlled by crypto. Factual · 28 Jul 2025 Technology
Resolved true Factual 27 Aug 2024 The Hunter Biden story was censored on the internet. Factual · 27 Aug 2024 Philosophy, ethics and society
Resolved true Factual 16 Aug 2024 Balaji Srinivasan has teamed up with Bryan Johnson to set up Blueprint-inspired food and fitness for the Network School community. Factual · 16 Aug 2024 Their own work or organisation
Mixed Factual 3 Aug 2024 Friedrich Hayek predicted Bitcoin. Factual · 3 Aug 2024 Economics and markets
Mostly true Factual Aug 2024 The Network School site in Malaysia is less than an hour from Singapore, which has an international airport (Changi). Factual · Aug 2024 Their own work or organisation
Mostly true Factual 22 Jul 2024 The US bureaucracy has tenure, so its officials cannot be removed by elections. Factual · 22 Jul 2024 AI policy and governance
Mostly true Factual 22 Jul 2024 America has $175 trillion in debt, which no election can pay off. Factual · 22 Jul 2024 Economics and markets
Mostly true Factual 7 Jun 2023 Balaji Srinivasan has created tweets, talks, podcasts, courses, articles and The Network State book. Factual · 7 Jun 2023 Their own work or organisation
Resolved true Factual 7 Jun 2023 There is a community chat associated with the newsletter at balajis.com/chat. Factual · 7 Jun 2023 Their own work or organisation
Resolved true Factual 7 Jun 2023 Balaji Srinivasan's newsletter audience has more than 100,000 subscribers. Factual · 7 Jun 2023 Their own work or organisation
Mostly true Factual 2023 Srinivasan conceded a bet and paid out about a million dollars, saying he had "burned a million to tell you they're printing trillions." Factual · 2023 Economics and markets
Mostly true Factual 4 Jul 2022 In the late 1990s, and probably the early 2000s, people (including Bill Gates) talked about a 'convergence device'. Factual · 4 Jul 2022 Technology
Mostly true Factual 4 Jul 2022 Inflation affects African Americans, Latinos, Asians and whites equally. Factual · 4 Jul 2022 Economics and markets
Resolved true Factual 4 Jul 2022 Sergey Brin remarked many years ago that staring down at a phone is emasculating. Factual · 4 Jul 2022 Technology
Mostly false Factual 17 Nov 2021 Under FDR, a broad interpretation of the Interstate Commerce Clause effectively repealed the 10th Amendment, leaving states without real rights. Factual · 17 Nov 2021 AI policy and governance
Mostly true Factual 17 Nov 2021 Miami has accepted some city coins and Texas is big in Bitcoin mining. Factual · 17 Nov 2021 AI policy and governance
Mostly true Factual 17 Nov 2021 The newspapers' coverage essentially got President Nixon forced out of office. Factual · 17 Nov 2021 Philosophy, ethics and society
Mostly false Factual 17 Nov 2021 22 Walgreens stores are closing in San Francisco because of shoplifting. Factual · 17 Nov 2021 Economics and markets
Mostly false Factual 11 Nov 2021 People share values with those close to them in their social network rather than with those close in physical space. Factual · 11 Nov 2021 Philosophy, ethics and society
Mostly true Factual 11 Nov 2021 The concept can be called 'dynamic digital geography', after a term Patri Friedman introduced and used for seasteading. Factual · 11 Nov 2021 Philosophy, ethics and society
Mostly false Factual 11 Nov 2021 Russia switched its ideology in 1991 from communism to nationalism. Factual · 11 Nov 2021 Philosophy, ethics and society
Mostly true Factual 11 Nov 2021 There is no public list of all Facebook and Twitter members. Factual · 11 Nov 2021 Technology
Resolved true Factual 11 Nov 2021 The Twitter and Facebook networks are each bigger than France or Germany combined. Factual · 11 Nov 2021 Technology
Mixed Factual 25 Mar 2021 The big media companies are basically nepotistic endeavors, passed down from father to son. Factual · 25 Mar 2021 Economics and markets
Mostly false Factual 25 Mar 2021 Cavalry became an outdated way of war that quickly went away after early World War I cavalry charges against machine gun nests. Factual · 25 Mar 2021 Other
Resolved true Factual 25 Mar 2021 Bitcoin's genesis block contains a reference to the bailouts. Factual · 25 Mar 2021 Economics and markets
Mostly true Factual 25 Mar 2021 Ethereum currently has a negative network effect: the more applications run on it, the higher the fees, which crowd out other uses. Factual · 25 Mar 2021 Technology
Mostly true Factual 31 Mar 2020 Peter Thiel had warned about biomedical red tape about five years before March 2020. Factual · 31 Mar 2020 Biosecurity
Mostly true Factual 30 Mar 2020 It took about 10 years for the economy to recover from the Great Recession. Factual · 30 Mar 2020 Economics and markets
Resolved true Factual 30 Mar 2020 Europe has reported low quality masks and tests imported from China. Factual · 30 Mar 2020 Biosecurity
Mixed Factual 29 Mar 2020 Vox did not accurately relay what experts were saying at the time; WHO had declared a global emergency in late January 2020, weeks before the Recode handshake piece. Factual · 29 Mar 2020 Other
Resolved true Factual 28 Mar 2020 Fear of retaliation may be why many doctors did not go on the record in recent weeks about equipment shortages. Factual · 28 Mar 2020 Biosecurity
Resolved true Factual 13 Mar 2020 If about 100 people in San Francisco had the virus, there would be about a 10 percent chance that a 1,000-person gathering includes an infected person. Factual · 13 Mar 2020 Biosecurity
Resolved true Factual 13 Mar 2020 Real-time PCR is not especially fast; 'real time' refers to something different. Factual · 13 Mar 2020 Biosecurity
Mostly true Factual 13 Mar 2020 COVID-19 testing involves taking a swab from the patient's mouth or nose, sending it to a lab, and receiving a positive or negative result. Factual · 13 Mar 2020 Biosecurity
Resolved true Factual 13 Mar 2020 COVID-19 is spreading uncontrollably on multiple continents, so it meets the definition of a pandemic. Factual · 13 Mar 2020 Biosecurity
Mostly false Factual 31 Dec 2019 Roughly two years of monetizing WhatsApp at Facebook's ARPU would yield about $19B in revenue, the amount Facebook paid for WhatsApp. Factual · 31 Dec 2019 Economics and markets
Resolved true Factual 26 Dec 2019 The Oculus Quest VR headset has been a sleeper hit commercially. Factual · 26 Dec 2019 Technology
Mostly true Factual 25 Dec 2019 A website with a .eth or other crypto domain can now hold a crypto wallet as well as metadata. Factual · 25 Dec 2019 Technology
Resolved true Factual 24 Dec 2019 PostgreSQL natively indexes JSON data via the JSONB type. Factual · 24 Dec 2019 Technology
Mostly true Factual 24 Dec 2019 As advertising load rises, customer satisfaction falls. Factual · 24 Dec 2019 Economics and markets
Resolved true Factual 25 Apr 2018 Software companies such as Andela are starting to emerge from Nigeria. Factual · 25 Apr 2018 Economics and markets
Mostly true Factual 25 Apr 2018 Proof-of-stake and delegated-proof-of-stake are consensus variants that work basically like shareholder votes. Factual · 25 Apr 2018 Economics and markets
Mostly true Factual 25 Apr 2018 It is possible to delete a Facebook account. Factual · 25 Apr 2018 Technology
Resolved true Factual 2 Apr 2018 With Ethereum and ICOs, startup funding has begun to decentralize. Factual · 2 Apr 2018 Economics and markets
Back to the list

Factual AI capabilities

“The current crop of agents can’t set complex goals, or properly verify outputs.”

Balaji Srinivasan, author of his own blog post 3 August 2025 AI is polytheistic, not monotheistic (balajis.com)

Current AI agents cannot set complex goals or properly verify outputs.

Mostly true True in substance, wrong in a detail.

What would count as true or false

Judge as of 2025-08-03, against the state of publicly available AI agents at that time. The claim says agents could not do two things: set complex goals, and properly verify their own outputs. TRUE: benchmark results, expert assessments or technical reports from mid-2025 show that agents of the time generally could not reliably set complex goals on their own and could not reliably verify their outputs (e.g. low success on long-horizon agentic benchmarks, documented failures at self-verification). MOSTLY TRUE: this holds broadly, but there are minor, narrow exceptions (e.g. verification in specific domains such as running code tests). MIXED: the evidence supports one of the two limitations but shows agents could do the other reasonably well. MOSTLY FALSE: agents of the time demonstrably did both in significant settings, so the blanket 'can't' overstates real but partial limitations. FALSE: agents of the time routinely and reliably did both.

Evidence

  1. Measuring AI Ability to Complete Long Tasks arXiv (METR authors: Kwa, West et al.)
    “We also noted important limitations of current systems, particularly their lower performance on less structured, “messier” tasks”
  2. Measuring AI Ability to Complete Long Tasks arXiv (METR authors: Kwa, West et al.)
    “On these tasks, current frontier AI models such as Claude 3.7 Sonnet have a 50% time horizon of around 50 minutes.”
  3. How Does Time Horizon Vary Across Domains? METR · 14 July 2025
    “The agent fails to click the correct coordinates despite planning detailed and accurate steps in their code comments, indicating strong planning but weak execution capabilities.”
  4. 4 Phases To Smarter AI Integration For Business Oliver Wyman · 22 July 2025
    “A benchmark study by Salesforce AI Research found that AI agents achieve only 58% success on single-turn tasks. Add multi-turn interactions, and performance drops to 35%.”
  5. Measuring AI Ability to Complete Long Tasks arXiv (METR authors)
    “On these tasks, current frontier AI models such as Claude 3.7 Sonnet have a 50% time horizon of around 50 minutes.”
  6. Measuring AI Ability to Complete Long Tasks arXiv (METR authors)
    “We also measure the 80% time horizon of models (Figure 6) and find a similar trend, though horizons are roughly 5x shorter.”
  7. Measuring AI Ability to Complete Long Tasks arXiv (METR authors)
    “We also note several limitations of current systems—notably, performance is much lower on less structured, “messier” tasks (Section 6.2).”
  8. How Does Time Horizon Vary Across Domains? METR · 14 July 2025
    “Agentic GUI tasks (OSWorld) are still ~100 × shorter than coding or math; Tesla self-driving has improved much slower than other domains.”
  9. On the self-verification limitations of large language models on reasoning and planning tasks ICLR 2025 Proceedings · 1 May 2025
    “We observe significant performance collapsewith self-critique and significant performance gains with sound external verification.”
  10. τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains arXiv (Sierra Research authors) · 2 May 2024
    “Our experiments show that even state-of-the-art function calling agents (like gpt-4o) succeed on \(<50\%\) of the tasks, and are quite inconsistent (pass^8 < 25% in retail).”
  11. τ2-Bench: Evaluating Conversational Agents in a Dual-Control Environment arXiv · 15 January 2025
    “gpt-4.1 pass \(\hat{}\) 1 drops from 74%/56% for retail and airline respectively to 34% for telecom.”
  12. System Card: Claude Opus 4 & Claude Sonnet 4 Anthropic · May 2025
    “They are particularly adept at complex computer coding tasks, which they can productively perform autonomously for sustained periods of time.”

Reasoning

An ICLR 2025 paper found that performance collapses when LLMs critique their own work on reasoning and planning tasks, and improves with sound external verification (E9). That supports the self-verification limitation. METR and other benchmarks show partial, unreliable capability on longer and messier tasks: a 50% time horizon of about 50 minutes, 80% horizons roughly 5x shorter, and low or inconsistent success on multi-turn tasks (E1, E2, E6, E7, E4, E10, E11). Together these show limited reliability on longer, less structured and multi-step tasks; they measure task completion rather than goal-setting, so they bear only indirectly on the claim that agents cannot set complex goals. There are narrow exceptions: Anthropic describes sustained autonomous coding (E12), and OSWorld agents planned well but executed poorly (E3). These temper the blanket "can't" without overturning it.

Cite this claim

Balaji Srinivasan, “Current AI agents cannot set complex goals or properly verify outputs.” (said 3 August 2025). Verdict: mostly true. Reliability API, claim c_87e91de51727, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_87e91de51727

Claims without a verdict

Is a verdict here wrong, or do you speak for Balaji Srinivasan? Reply on the claim's own page, with the form at its foot (how replies work). Verdicts are made automatically; how verdicts and scores are made.