E166: Mind-blowing AI Video: OpenAI launches Sora + Is Biden too old? Tucker/Putin interview & more

Chamath’s comment came during a discussion of AI startups raising $100M rounds, much of which was being spent on GPU compute. He said that by "next week" (i.e., roughly late February 2024) "when all these announcements are done ... you're about to see a one‑tenth of the compute cost," framing it as going from spending $70–80M of a $100M round on compute to spending $7–8M for the same throughput. (podscripts.co)

In the very next episode (E167 on Feb 23, 2024), one of the main segments is explicitly titled "Groq's big week, training vs. inference, LPUs vs. GPUs," indicating that the "announcements" he was previewing were about new inference hardware like Groq’s Language Processing Unit (LPU). (metacast.app) Just before and during that week, Groq’s LPU was publicly benchmarked and heavily covered: a Feb 13, 2024 press release and independent benchmarks showed its LPU inference engine delivering dramatically higher throughput (hundreds of tokens/sec) than existing GPU-based providers on Llama 2‑70B, marketed as a 10x step-change in speed and efficiency. (groq.com)

On February 22, 2024—within the "next week" window—Beebom and other outlets summarized Groq’s own claims that its LPU clusters could perform LLM inference 10× faster and at roughly 1/10th the cost per token compared to Nvidia H100 GPU clusters, explicitly using the “10x speed” and “1/10th the cost” language Chamath had alluded to. (beebom.com) Follow‑up technical coverage (e.g., The Next Platform) clarified that the “1/10th cost” is best understood in terms of per‑token/time/energy cost for inference rather than literal sticker price of full systems, but it still represents an order‑of‑magnitude reduction in effective compute cost for those workloads relative to the prevailing GPU setups. (nextplatform.com)

By contrast, there was no immediate, industry‑wide 10× drop in GPU prices or in the overall cost of AI compute in that same week—H100s remained extremely expensive, and Nvidia’s next major efficiency jump (Blackwell, announced March 18, 2024) was future‑dated and did not instantly cut users’ bills by 10×. (investors.com) So if you interpret his remark as “the whole market’s compute costs will suddenly be 10× lower next week,” it would be wrong. But taken in context—previewing upcoming announcements about new inference hardware that offers ~10× better cost‑per‑throughput versus the then‑standard Nvidia GPU stacks—the prediction that such announcements were imminent was essentially borne out by Groq’s LPU launch coverage and benchmarks in that exact time window.

Given the format of your normalized prediction (“announcements will be made that reduce the cost of AI compute (for comparable throughput) by roughly 10x versus the then‑prevailing cost structure”), and the fact that those Groq announcements with 10x cost‑and‑speed claims did occur in the following week, the most reasonable classification is that the prediction was right, albeit narrowly (it applied to specific inference offerings, not to all AI compute).

Friedberg's claim was that the next major iPhone and flagship-Android cycle after February 2024 would ship with chips explicitly designed to run AI models locally and enable substantial on‑device AI features. That is what happened.

Apple's iPhone 16 lineup with A18 and A18 Pro chips (launched September 2024) is marketed as being built for Apple Intelligence; Apple states that many of the generative models behind Apple Intelligence run entirely on‑device, and that third‑party apps can call those on‑device models via the Foundation Models framework, with features that can work offline. (apple.com) Apple and press materials also describe the A18 Neural Engine as optimized for large generative models and significantly faster at running ML workloads, specifically to accelerate Apple Intelligence, and note that Apple Intelligence requires newer A‑series chips with sufficient GPU/NPU capacity for on‑device processing. (businesswire.com)

On Android, Samsung's Galaxy S25 series uses Qualcomm's Snapdragon 8 Elite for Galaxy, whose substantially faster NPU is highlighted as enabling many more Galaxy AI experiences to run on‑device, including features like Generative Edit that previously relied on the cloud; Samsung also exposes a setting to disable cloud AI and rely only on local processing. (techpowerup.com) Google's Pixel 9 family, powered by the Tensor G4 chip, runs Google's Gemini Nano large language model locally; this on‑device LLM powers features such as Recorder summaries, Call Notes, Pixel Screenshots, and other real‑time summarization, translation, and captioning capabilities that operate without sending data to the cloud. (androidauthority.com) Other flagship SoCs like MediaTek's Dimensity 9400 and Samsung's Exynos platforms are likewise advertised as supporting on‑device generative AI, including multimodal Gemini Nano, underscoring that high‑end phones are now expected to run such models locally. (semiconductor.samsung.com)

Given that the post‑February‑2024 flagship cycles from Apple, Samsung, Google, and other major vendors did in fact ship with dedicated NPUs/AI engines specifically marketed for running sizable generative models on the device and enabling prominent on‑device AI features, Friedberg's prediction that the 'next set of mobile devices' would have chips that run models locally has been borne out.

Chamath’s claim bundles several linked predictions: (1) the first major real‑world impact of Sora‑like AI video tools will be in pornography, (2) synthetic, non‑human porn will become dominant, and (3) this will “destroy pornography in OnlyFans” / significantly erode human‑creator platforms.

On point (1), there is some support but not a clean confirmation:

• OpenAI’s own Sora products explicitly ban graphic sexual content: prompts requesting sexual/nude material are blocked and outputs are filtered, making official Sora a poor vehicle for mainstream porn. (pureai.com)
• However, other AI video tools targeted at adult content have grown quickly. Industry pieces and market reports suggest AI‑generated porn is already a substantial minority of new content: around 30% of new adult content in 2025 on major platforms, with similar estimates for AI’s share of new uploads and LGBTQ+ segments. (reelmind.ai) Broader analyses estimate 15–20% of new adult content consumed in 2025 involves AI (deepfakes, generators, AI chat/companions). (whatstrending.com) Other market data show rapid AI adoption in the adult sector and a projected AI‑porn market of about $2.4B by 2026, with high adoption of synthetic performers and virtual companions. (gitnux.org)
  This suggests the adult industry is indeed one of the earliest and heaviest adopters of generative video, which partially tracks the “you’ll see it here first” intuition, but it’s not exclusive to porn and is hard to prove as the single first/primary impact domain.

On points (2) and (3), the core outcomes have clearly not materialized yet:

• Even optimistic, industry‑friendly sources put AI‑generated porn at roughly a large minority (≈15–30%) of new content, not the majority; human‑performed porn still dominates total supply and consumption. (reelmind.ai)
• Data from OnlyFans show continued strong growth rather than erosion. Official filings and analyses report:
  • 2023 revenue around $1.3B with nearly 20% year‑over‑year growth and billions paid out to creators. (upmarket.co)
  • For 2024, user accounts up 24% to ~377.5M, creators up to ~4.6M, total payments up to $7.2B, with pre‑tax profits and owner dividends hitting record highs ahead of a possible ~$8B sale. (businessinsider.com)
    There is no evidence that AI porn has “destroyed” or even measurably weakened OnlyFans’ core human‑creator business so far.
• Some subscription competitors (e.g., Fanvue) do lean into AI‑generated models, and at one point AI creators made about 15% of Fanvue’s revenue, showing a real but still partial shift toward synthetic performers on certain platforms. (en.wikipedia.org) This indicates competitive pressure and experimentation, not wholesale replacement.

Because Chamath explicitly framed the outcome as happening “over time”, and we are only about 21 months past the February 2024 prediction, it is too early to say whether synthetic porn will ultimately become dominant or whether it will meaningfully undercut human‑based platforms in the longer term. Current evidence shows:

• AI porn is growing fast and is an important early use‑case for generative video.
• It has not yet become the dominant form of porn.
• Human‑creator businesses like OnlyFans are, so far, growing rather than collapsing.

Given the vague time horizon and mixed signal—some early support for the “porn will adopt this first” intuition, but clear non‑realization of the stronger claims about dominance and destruction—the fairest assessment as of November 30, 2025 is **“inconclusive (too early)” rather than definitively right or wrong.

Nvidia’s valuation has grown dramatically since the prediction, but it has not reached $10T, nor has anything happened that clearly makes a $10T outcome impossible.

• At the time of the podcast (February 2024), Nvidia was crossing the ~$2T mark; news coverage notes it hit $2T in February 2024 and then rose to about $2.25T by March 2024, up from roughly $1.2T at the end of 2023. (ndtv.com)
• As of late November 2025, Nvidia’s market cap is around $4.4T, making it the world’s most valuable company, and it has traded as high as roughly $4.5T in October 2025. (companiesmarketcap.com)
• The condition attached to the prediction—continued strength in AI hardware/efficiency trends—largely holds so far: Nvidia remains the dominant AI chip supplier, Blackwell-based GPUs are ramping strongly, and the roadmap to Rubin and beyond underpins very high expected AI demand and revenue (management and analysts discussing targets of hundreds of billions in AI-chip revenue through 2026). (cnbc.com)

However, the prediction was long term and framed as potential (“could be worth $10 trillion, potentially”), with no specific deadline. Nvidia has not yet reached $10T, but given that it is already in the mid‑trillions and AI hardware demand remains elevated, it is not possible to say that a $10T valuation has been disproven. It remains an open, speculative outcome rather than a resolved one.

Because the core claim is about a possible long‑term ceiling and no explicit timeframe has elapsed in which that ceiling would need to be hit or clearly ruled out, the status of “Nvidia could be worth $10T in the long term” is too early to judge.

There is strong evidence that the tooling part of Chamath’s claim has materialized, but little hard evidence that it has already transformed non‑English‑speaking workers into core, top‑productivity startup contributors at scale.

Why it looks directionally plausible / partially true

• Major collaboration platforms now provide near real‑time spoken translation, removing a large part of the language barrier in meetings. Google Meet and DeepMind launched real‑time speech translation features in 2025, translating spoken language during calls into participants’ preferred languages, with support for multiple European languages. (blog.google) Cisco is acquiring EzDubs to add voice‑preserving real‑time translation into Webex, explicitly to let people converse across languages. (indianweb2.com) Similar capabilities exist across specialized meeting tools like Interactio, Wordly, and others marketed for business meetings, trainings, and events. (wordly.ai)
• General‑purpose AI systems such as OpenAI’s GPT‑4o support real‑time multilingual voice conversations and translation in over 50 languages, explicitly pitched as removing language friction in everyday and professional use. (lifewire.com) Dedicated products (e.g., Transync AI, emotii.ai, Relay’s TeamTranslate) target workplaces and multinational teams, advertising that they "eliminate language barriers" and let every worker be "heard, understood, and empowered." (apps.apple.com)
• In adjacent high‑skill domains there is empirical evidence that generative AI is disproportionately helping non‑English speakers. A 2025 study of 2M+ biomedical papers finds AI‑assisted writing adoption has grown about 400% in non‑English‑speaking countries (vs. 183% in English‑speaking ones), modestly increasing productivity and narrowing publication gaps—clear evidence that AI is mitigating English language disadvantages in knowledge work. (arxiv.org) Research prototypes like WordDecipher specifically target digital workplace communication for non‑native English speakers, helping them craft socially appropriate English messages in email and chat. (arxiv.org)

Why we still can’t classify the prediction as clearly right or wrong

• Chamath’s prediction is not just that tools exist, but that they “basically become some of your best folks”—i.e., that non‑English speakers, relying on AI to bridge language, are now functioning as core, highly productive startup/tech team members at a level comparable to English‑fluent peers.
• The available evidence shows rapid deployment of translation and language‑assist tools and early proof that they reduce language friction and help non‑English professionals (scientists, remote workers, call‑center agents, etc.). (nypost.com) However, there are no large‑scale labor‑market studies or broad startup case‑study datasets demonstrating that non‑English‑speaking workers who rely primarily on AI (rather than learning English) are now systematically regarded as top, core contributors in startups or tech firms.
• Many startups still explicitly require English proficiency in job postings, and qualitative reporting focuses more on potential and early pilots than on a documented, widespread shift in team composition and top‑performer profiles.

Because the mechanism (AI offloading language) is clearly emerging but the outcome (non‑English speakers broadly becoming some of the best core team members thanks to these tools) isn’t yet measurable or well‑documented, the prediction can’t be cleanly scored as right or wrong with current evidence—hence ambiguous.

The prediction was that Joe Biden would not voluntarily drop out of the 2024 presidential race and would run through the November 2024 election.

In reality, Biden ended his reelection bid on July 21, 2024, announcing in a public letter that he would "stand down" from the campaign and not seek reelection, and that he would instead focus on serving out the remainder of his term while endorsing Kamala Harris as the Democratic nominee. (dallasnews.com) Harris went on to become the nominee, and Biden was not on the ballot in November 2024. (en.wikipedia.org)

Although there was substantial pressure from Democratic leaders and donors for him to step aside, the literal outcome that matters for evaluating the prediction is that he did, in fact, withdraw and did not run in the November election, directly contradicting the forecast that he would "go for it" and stay in the race through Election Day.

Therefore, the prediction is wrong.

Factual outcome:

• Joe Biden did secure enough delegates in the 2024 Democratic primaries and became the presumptive nominee, winning an overwhelming majority of pledged delegates. (en.wikipedia.org)
• After a widely criticized June 2024 debate and intense intra-party pressure, Biden withdrew from the 2024 race on July 21, 2024, becoming the first eligible incumbent since Lyndon B. Johnson in 1968 to end a reelection bid after locking up the delegates. (en.wikipedia.org)
• Following his withdrawal, Biden endorsed Kamala Harris, who was then chosen by the delegates and became the official Democratic nominee on August 5, 2024. (en.wikipedia.org)
• Harris, not Biden, was on the ballot as the Democratic nominee in November 2024. (en.wikipedia.org)

Interpretation vs. prediction:

• The normalized prediction: “Barring an extreme unforeseen event (e.g., major health crisis), Joe Biden will secure enough primary delegates and will be the official Democratic nominee for U.S. president in the 2024 election.”
• Reality: Biden did secure enough primary delegates, but he did not become the official nominee, because he later withdrew and declined the nomination.
• However, Sacks explicitly framed his forecast as conditional: “unless he changes his mind or there’s some act of God, he’s going to be the nominee.” Biden did change his mind and stepped aside in what contemporaneous coverage and later analysis treat as an unprecedented, extraordinary political event—an incumbent withdrawing after clinching the delegates under intense pressure from his own party. (en.wikipedia.org)

Because the prediction was explicitly limited to the world “barring an extreme unforeseen event / unless he changes his mind,” and the actual world is precisely one in which he changed his mind amid an extraordinary, historically rare chain of events, we can’t cleanly score the statement as right or wrong. The outcome falls outside the scenario he was conditioning on.

Hence, the fairest evaluation is ambiguous rather than definitively right or wrong.

As of mid‑February 2024, the major declared presidential candidates included Joe Biden, Donald Trump, Robert F. Kennedy Jr., Cornel West, Jill Stein, Dean Phillips and others—but not Kamala Harris as a presidential candidate. Biden was running for reelection with Harris as his vice‑presidential running mate, not as a separate presidential contender. (en.wikipedia.org)

On July 21, 2024, Biden announced he would not seek reelection and endorsed Harris to replace him at the top of the Democratic ticket. This announcement triggered the launch of the Kamala Harris 2024 presidential campaign and a rapid consolidation of party support around her. (reuters.com) Harris’s campaign is widely dated from this July 21 entry into the race and is noted for being only 107 days long, the shortest general‑election presidential campaign in modern history. (en.wikipedia.org)

Harris quickly became the Democratic presidential nominee, securing the nomination via a delegate roll call in early August 2024 and heading the party’s national ticket against Donald Trump. (en.wikipedia.org) In the November 5, 2024 general election, she won 48.3% of the popular vote and 226 electoral votes, losing narrowly to Trump (49.8% and 312 electoral votes). That level of vote share, Electoral College support, and national prominence clearly qualifies her as a serious contender in the 2024 election. (en.wikipedia.org)

Because Harris was not a presidential candidate as of mid‑February 2024, but later entered the race and became the Democratic nominee and a major general‑election contender, her candidacy is a direct counterexample to Sacks’s prediction that no such new major presidential candidate would successfully enter and become a serious contender in 2024.

Joe Biden (whose ticket became the Democratic Harris–Walz ticket) and Donald Trump (with JD Vance) appeared on the presidential ballot nationwide as the major-party nominees, covering all 50 states and the District of Columbia. (en.wikipedia.org) Robert F. Kennedy Jr., however, did not achieve nationwide ballot access. A CBS News ballot map and later compiled results show that he was on the ballot only in a subset of states totaling about 309 electoral votes, far short of all 538 electoral votes available from 50 states plus DC. (cbsnews.com) Kennedy was explicitly removed or failed to qualify in states such as New York and Georgia, among others, and litigation confirmed his disqualification from the New York ballot. (reuters.com) His campaign also voluntarily withdrew from several state ballots after he ended his independent run and endorsed Trump, further limiting his ballot presence. (en.wikipedia.org) Because RFK Jr. was not on the ballot in all 50 states, the prediction that Biden, Trump, and Kennedy would all appear on every state’s November 2024 general-election ballot is incorrect.

Evidence indicates that a new, explicitly RFK‑linked third party has in fact been formally (re)created after the November 2024 election, using the ballot‑access infrastructure built for his independent run.

• During the 2024 cycle, Robert F. Kennedy Jr. created We The People Party (plus the Texas Independent Party) in multiple states specifically as a ballot‑access vehicle for his independent presidential campaign, not as a long‑term programmatic party.(en.wikipedia.org) This is the campaign infrastructure Chamath was referring to.
• In 2025, Kennedy supporters formally relaunched We The People Party as a standalone national third party, with a national chair (Levi Leatherberry), explicit plans to expand ballot access over the next three years, target races like the New York governorship, endorse candidates (e.g., Larry Sharpe for NY governor), and position the party as an “anti‑establishment” alternative for disaffected voters.(yahoo.com) This relaunch happened after the 2024 presidential election and is explicitly described as a way to reuse the party infrastructure built for Kennedy’s 2024 ballot‑access efforts.
• While the party is still small and its long‑term success is uncertain, it meets the core criteria of Chamath’s prediction: a formally organized, continuing third party, built out of RFK Jr.’s 2024 ballot‑access machinery, created after the presidential election with the stated intent to contest future midterms and the 2028 presidential race.(yahoo.com)
• By contrast, No Labels ended its 2024 presidential effort before the election and has not itself become a new, durable national party (its main structure remains a 501(c)(4) advocacy group, with only scattered state‑level party lines and even some loss of party recognition in places like North Carolina).(en.wikipedia.org)

Given that the central event Chamath described—post‑election creation of a continuing third party leveraging RFK’s 2024 ballot‑access infrastructure—has occurred, the prediction is best classified as right, even though the eventual durability and scale of We The People Party remain to be seen.

According to certified 2024 general-election results, Robert F. Kennedy Jr. ran as an independent and received roughly 0.48–0.49% of the national popular vote. His best statewide performance was about 1.96% of the vote in Montana, and in all other states his share was lower; no state shows him at or above even 2%, let alone 30%. (en.wikipedia.org)

Summary tables of third‑party and independent candidates for the 2024 presidential election explicitly list Kennedy’s “best state percentage” as 1.96%, confirming that he did not reach 30% in any state. (en.wikipedia.org)

Kennedy also suspended his independent campaign and endorsed Donald Trump in August 2024, further limiting his eventual vote share. (reuters.com)

Because he never came close to 30% of the popular vote in any state in the November 2024 election, Chamath’s prediction that RFK Jr. would get “30 plus percent” support in some states was not borne out by the actual results.