The Big Blob of Compute Hypothesis
StrategyProgress is driven primarily by raw compute, data quantity, data quality, and training duration rather than clever new methods.
3 more supporting quotes are in the member records.

다리오 아모데이
Co-founder and CEO of Anthropic
535 records in all
3 free questions a day, no sign-up.
35 basic records of 231
Principles are our summary of their public statements.
Progress is driven primarily by raw compute, data quantity, data quality, and training duration rather than clever new methods.
3 more supporting quotes are in the member records.
A breakthrough in the lab does not immediately translate to market revenue because of regulatory, manufacturing, and adoption lags.
Supporting quotes
If you train at a small context length and then try to serve at a long context length, maybe you get these degradations.2 more supporting quotes are in the member records.
Do not spend massive capital just because it sounds cool or competitors are doing it; every major investment must be backed by a detailed spreadsheet and risk assessment.
3 more supporting quotes are in the member records.
A business must allocate a portion of its resources to serving customers to generate the revenue needed to fund future research and development.
3 more supporting quotes are in the member records.
When markets fail to deliver technology to developing regions, proactive collaboration with philanthropists is necessary to ensure equitable access.
When the exact nature of a risk is unclear, focus on disclosing procedures and data rather than setting rigid rules that might become obsolete or burdensome.
Supporting quotes
Often this is for good reasons: governments have grave powers, and it's usually for the best that they aren't used too hastily. But the mismatch in timescale is nevertheless very painful2 more supporting quotes are in the member records.
To maintain a fair balance of power, individuals and organizations should have access to AI tools that are at least as capable as those used by the state or regulatory bodies against them.
Supporting quotes
The last thing we want is for the benefits of AI to be slowed while its risks loom large, so it's important to take action on this problem as soon as possible.2 more supporting quotes are in the member records.
Success in high-stakes technology comes from forming coalitions that share benefits and supply chains with those who align with common values.
3 more supporting quotes are in the member records.
When discussing the future, it is better to provide a semi-analytical assessment of progress rather than abstract or purely speculative guesses.
3 more supporting quotes are in the member records.
Focus on broad measurement tools or techniques that allow for programmable intervention, as a few such discoveries drive the majority of progress.
3 more supporting quotes are in the member records.
Major breakthroughs often come from creative connections and scrappy projects rather than massive funding or random search.
3 more supporting quotes are in the member records.
Leaders should increase scientific understanding to prevent a gap where those who need decision-making tools most are the ones who opt out of them.
Supporting quotes
One hopeful sign is that historically anti-technology movements have been more bark than bite: railing against modern technology is popular, but most people adopt it in the end, at least when it's a matter of individual choice.Human conflict is adversarial and AI can in principle help both the "good guys" and the "bad guys". If anything, some structural factors seem worrying: AI seems likely to enable much better propaganda and surveillance1 more supporting quote is in the member records.
Leaders should focus on using new technologies to actively improve institutions, rather than just defensively responding to risks.
3 more supporting quotes are in the member records.
Fast-paced innovation quickly turns luxury goods into affordable commodities, leveling the playing field for all users.
Founders should have explicit conversations with investors about non-traditional governance structures to ensure everyone understands potential conflicts with shareholder value.
Build an institution that is judged by its incentives and merits rather than the charisma of a single leader.
Companies should compete on safety and reliability rather than just speed or raw power to ensure long-term viability.
Supporting quotes
pacing does not mean halting model training or technical progress, but ensuring companies take adequate time to align and safeguard their models, and for third party evaluators to confirm this.2 more supporting quotes are in the member records.
Allow external reviewers to publish findings without editorial control, even if the results are unfavorable, to build genuine public trust.
Supporting quotes
Embedding evaluators may sound like a small or inconsequential step, but often the things that sound most boring or procedural are actually the most essential.Embedded evaluators can check at the level of nuts and bolts whether an AI company is actually following the training, deployment, operational, and safeguards practices they claim to be following.1 more supporting quote is in the member records.
Any cooperation or 'pause' in development must have ironclad verification to ensure competitors do not secretly defect and gain a dominant advantage.
3 more supporting quotes are in the member records.
Models have nuanced properties and behaviors that benchmarks cannot capture, requiring ongoing observation similar to getting to know a human.
Supporting quotes
Any kind of scheme you come up with, the reality tends to frustrate that scheme, right? It tends to break out of the scheme.2 more supporting quotes are in the member records.
You are seeing 81 basic records. Membership answers from all 535 of this master's records.
See membershipAI and frontier tech
These records are our compilation of public statements and sources. We are not affiliated with or endorsed by Dario Amodei or any related company. If something should come down, tell us. Takedown request form