E174: Inflation stays hot, AI disclosure bill, Drone warfare, defense startups & more

The prediction is time‑bound to “by the end of 2025”, and today’s date is December 1, 2025, so the deadline has not yet passed.

Public data on the AI compute market in 2025 shows that NVIDIA still overwhelmingly dominates deployed AI inference hardware (over ~80% of deployed inference GPUs), with AMD, Intel and others taking most of the remaining share, while Groq is described as a fast‑growing but still relatively small specialist provider. (sacra.com) Recent coverage of Groq emphasizes ambitious plans and targets—such as deploying ~100,000+ LPUs and aiming at up to 50% of global inference capacity—but these are framed as forward‑looking projections tied to large contracts (e.g., Saudi/Aramco) rather than as already‑achieved market share, and no credible source reports Groq as anywhere near 50% of global inference compute today. (ainvest.com)

Given that (a) the explicit horizon (end of 2025) has not yet arrived, and (b) current evidence strongly suggests the prediction is very unlikely to be met but not yet logically impossible before December 31, the status must be classified as “inconclusive (too early)” rather than definitively right or wrong.

Evidence from Federal Reserve meeting records shows that the federal funds target range stayed at 5.25%–5.50% at every FOMC meeting from January through June 2024, with no interest rate cuts implemented before July 2024. The January 31 and March 20, 2024 meetings both left the benchmark rate unchanged at 5.25%–5.50%. (pcbb.com)

The April/May 1, 2024 and June 12, 2024 FOMC meetings likewise held the rate steady at 5.25%–5.50%, explicitly choosing not to cut despite earlier market expectations for reductions. (federalreserve.gov) A consolidated calendar of FOMC decisions confirms that from January 31 through June 12, 2024, every meeting result was “Held Steady” at 5.25%–5.50%. (proinvestnews.com)

At the July 31, 2024 meeting, the Fed again maintained the federal funds rate in the same 5.25%–5.50% range, meaning there was still no cut as of that date. (federalreserve.gov) Subsequent commentary around the June decision noted that most forecasters were ruling out a July cut and looking instead to later in 2024, which is consistent with the idea that any eventual cut would come in July or later, not earlier in the year. (cbsnews.com)

Since (1) there were no interest rate cuts before July 2024, and (2) when cuts were discussed, they were framed as happening in July or later rather than earlier in 2024, Sacks’s prediction that the Fed would not implement cuts before July and would delay them until July 2024 or later is borne out by the actual policy path.

• Condition of the prediction was not met. Sacks’ prediction was explicitly conditional: “if we see a rate hike before the election, I think Biden is toast.” In reality, the Federal Reserve did not raise the federal funds rate at any meeting between the podcast date (April 12, 2024) and the November 5, 2024 election. The Fed held the target range at 5.25%–5.50% through mid‑2024 and then cut rates to 4.75%–5.00% on September 18, 2024, rather than hiking.(federalreserve.gov) Because the rate‑hike condition never occurred, the prediction concerns a counterfactual scenario.

• Election outcome: In the actual 2024 U.S. presidential election on November 5, 2024, Republican nominee Donald Trump defeated Democratic nominee (and sitting vice president) Kamala Harris; the presidency shifted from Democrat Joe Biden’s administration to Trump, a Republican.(en.wikipedia.org) So Biden and his party did lose control of the White House—but this happened without the antecedent (a pre‑election Fed rate hike) ever taking place.

• Why this is labeled ambiguous rather than right or wrong: The forecast was about what would happen if the Fed hiked before the election. Since that condition never materialized, we have no empirical way to test whether a rate hike would, in fact, have doomed Biden’s reelection prospects. Treating the statement as true or false would require interpreting it as an unconditional claim about the election, which is not what was actually said. Therefore, even though the Democrats ultimately lost the presidency, the specific conditional prediction can’t be definitively judged as correct or incorrect and is best classified as ambiguous.

Evidence on realized inflation matches the normalized prediction.

• The Fed’s 2% target is defined in terms of the PCE price index. Estimates from the St. Louis Fed and other analyses show that headline PCE inflation for calendar year 2024 was about 2.6%, i.e., clearly above 2%.
• Monthly data from BEA releases and compiled series indicate that the 12‑month change in headline PCE stayed above 2% in every month of 2024 (roughly 2.6% in January, drifting down to about 2.1–2.4% by the fall, and back up to ~2.6% in December). It never touched or went below 2%.
• CPI inflation, while not the Fed’s target, also reinforces the idea of “persistent” inflation: headline CPI ran between about 2.4% and 3.5% year‑over‑year throughout 2024, never falling to 2.0%.
• Into early 2025, headline PCE remained around 2.5–2.7% year‑over‑year, still above target, confirming that disinflation back to 2% did not materialize in the near term after his April 2024 comment.

Given that inflation (on the Fed’s preferred PCE measure) stayed persistently above 2% through the end of 2024 and beyond, the normalized prediction that a quick return to 2% was unlikely is borne out by the data.

The prediction was explicitly conditional:

If you don’t see [inflation] change in the next three months, you’re going to see 75/25 for a hike of at least 25 basis points.

That is, only if U.S. inflation dynamics failed to improve materially between roughly April 11 and mid‑July 2024 did Chamath expect market‑implied odds to move to about 75% for a Fed rate hike (vs. a cut).

1. What actually happened to inflation (the condition)

• March 2024 CPI was hot: headline 0.4% m/m and 3.5% y/y; core CPI 0.4% m/m and 3.8% y/y, with the 3‑month annualized core rate around 4.5%, prompting concern that inflation was re‑accelerating. (cnbc.com)
• From April through June 2024, inflation clearly cooled:
  • April: CPI up 0.3% m/m and 3.4% y/y; core 0.3% m/m and 3.6% y/y, the lowest core reading since April 2021. (cnbc.com)
  • May: headline CPI flat m/m (0.0%) and 3.3% y/y; core 0.2% m/m and 3.4% y/y, also the lowest annual core since April 2021. (cnbc.com)
  • June (reported July 11): CPI fell −0.1% m/m and 3.0% y/y, its lowest in over a year; core rose just 0.1% m/m and 3.3% y/y, the smallest annual core increase since April 2021. Several analyses noted that over the prior three months, inflation was running at about a 1% annualized pace, emphasizing a renewed disinflation trend. (cnbc.com)

This is a material improvement in inflation dynamics over the three months after his statement: sequential slowing in April and May, then outright negative monthly CPI and very soft core in June. The antecedent of his conditional (“if you don’t see this thing change in the next three months”) therefore did not occur.

2. What actually happened to market‑implied odds by mid‑July 2024

Even though the condition failed, we can look at real‑world pricing:

• After the soft June CPI release on July 11, 2024, CME FedWatch data (via CNBC) showed markets pricing >80% odds of a September rate cut, with investors increasing, not decreasing, cut expectations. (cnbc.com)
• Other coverage just before and around that time similarly reported FedWatch probabilities showing a strong majority chance of cuts by September and essentially no market focus on hikes. (benzinga.com)

So in the actual path of the world, odds moved in the opposite direction of his described 75%‑for‑a‑hike scenario—but that was in a world where inflation did improve.

3. Why the outcome is scored as ambiguous

Chamath’s forecast is about a counterfactual: what Fed‑funds futures would look like if inflation had stayed hot and not “changed” over the next three months. In reality, inflation cooled materially, so the condition attached to his prediction was false. Forecast‑evaluation frameworks typically treat such conditional predictions as not testable when their condition doesn’t occur.

• We can say that the world he described (high hike odds) did not match the world we got, but we cannot infer from that whether he was right or wrong about what markets would have done under his specified “no improvement in inflation” scenario.

Because the antecedent of the conditional failed, and the statement was explicitly conditional, the correctness of his specific claim about “75/25 for a hike” cannot be determined from realized data.

Hence the appropriate label is "ambiguous" rather than right or wrong.

The prediction specifies a time window for achieving “some form of AGI”: roughly 5–10 years from 2024, i.e., between 2029 and 2034. That window has not started yet as of the evaluation date (December 1, 2025); the earliest point, 2029, is still in the future.

While there is active debate about how close current large models (like cutting‑edge GPT-based systems and other frontier models) are to AGI, there is no consensus or widely accepted evidence that AGI has already been achieved as of 2025. However, this doesn’t invalidate the prediction, because it was about future achievement by 2029–2034, not by 2025.

Since the specified time range has not elapsed (and has not even begun at the early bound), the correctness of the prediction cannot yet be determined. Therefore, the status is inconclusive (too early to tell).

As of December 1, 2025, the key end‑states in Sacks’s prediction have not happened yet, but the process he described is clearly underway.

What has happened:

• Multiple major lawsuits over AI training on copyrighted content have been filed against AI companies (e.g., Thomson Reuters v. Ross Intelligence; Bartz v. Anthropic; Kadrey v. Meta; New York Times v. OpenAI). (reuters.com)
• U.S. district courts have issued conflicting decisions on fair use in the AI‑training context:
  • In Thomson Reuters v. Ross Intelligence (D. Del.), the court held that using Westlaw headnotes to develop an AI legal‑research tool was not fair use and granted summary judgment for Thomson Reuters. (reuters.com)
  • In Bartz v. Anthropic (N.D. Cal.) and Kadrey v. Meta (N.D. Cal.), judges held that using lawfully obtained books to train large language models is highly transformative and can qualify as fair use on the records before them, even while distinguishing or condemning uses of pirated copies. (theguardian.com)
    These are exactly the kind of “messy” fair‑use fights Sacks foresaw, but they are still at the trial‑court level.
• Appeals are pending (for example, Thomson Reuters v. Ross is now before the Third Circuit as case 25‑2153, and Bartz v. Anthropic has spawned a Rule 23(f) petition and Ninth Circuit amicus briefs), but no U.S. court of appeals has yet issued a definitive generative‑AI‑training fair‑use ruling that creates a clear circuit split. (mishcon.com)

What has not happened yet (and is central to the prediction):

• No U.S. Supreme Court decision has resolved whether training generative‑AI models on copyrighted works is fair use. Current Supreme Court copyright jurisprudence (e.g., Andy Warhol Foundation v. Goldsmith in 2023) addresses transformative use generally, not AI training, and recent coverage of the Anthropic ruling explicitly notes that legal experts still expect the Supreme Court may eventually have to address AI training. (en.wikipedia.org)
• No AI‑training‑specific federal copyright statute has been enacted. Bills such as the Generative AI Copyright Disclosure Act and the bipartisan TRAIN Act have been introduced and referred to committee, but as of late 2025 they have not passed Congress. Other enacted laws (like the TAKE IT DOWN Act and proposed NO FAKES Act) focus on deepfakes and likeness rights, not on whether AI training itself is fair use. (en.wikipedia.org)

Because Sacks’s full sequence—circuit‑level split → Supreme Court fair‑use ruling on AI training → follow‑on act of Congress—has not yet occurred, but the timeline (“in the coming years”) still allows for it and the litigation/legislative machinery is clearly in motion, the prediction cannot yet be judged right or wrong. It is therefore too early to call, i.e., inconclusive rather than clearly correct or clearly false.

By late 2025, text‑to‑music systems like Suno and Udio have become widely used by both creators and consumers: Udio reports millions of users since its 2024 launch, its tracks have gone viral and even entered national charts (e.g., an Udio‑generated song reaching the German Top 50), and Suno‑based songs have driven major TikTok/streaming controversies and licensing deals with the three major labels. (cirrkus.com) AI‑generated artists and songs are also charting commercially (for example, an AI gospel artist topping Christian and Gospel digital charts and the viral, fully AI‑generated song “We Are Charlie Kirk” topping Spotify’s viral songs chart), indicating exactly the kind of prolific use by musicians and ordinary users Friedberg described. (nypost.com) On the legislative side, Tennessee enacted the ELVIS Act in 2024, explicitly framed as the first U.S. law to protect musicians from AI voice cloning and audio deepfakes, directly targeting generative music/voice tools. (en.wikipedia.org) In Congress, lawmakers introduced the No AI FRAUD Act to create a federal right over one’s voice and likeness against AI fakes—promoted by its sponsors and the RIAA as necessary to protect music artists from AI‑driven impersonation—and additional bills like the TRAIN Act seek transparency around unauthorized AI training on creators’ works. (salazar.house.gov) Abroad, the EU’s AI Act and national initiatives such as Denmark’s proposed law granting people rights over AI‑generated imitations of their body, face and voice, plus UK parliamentary committees explicitly focusing on AI’s use of copyrighted music (accompanied by a silent protest album by 1,000 artists), show legislators actively grappling with generative‑AI impacts on music and related rights. (en.wikipedia.org) Taken together, rapid mainstream adoption of generative music tools and the emergence of music‑focused or music‑driven AI laws and bills match Friedberg’s forecast that widespread use would force legislators to confront and debate regulation of generative music AI, so the prediction is essentially correct by late 2025.

Sacks’ claim was conditional and medium‑ to long‑term: if cheap drones/missiles kept being used against U.S./allied ships while the U.S. remained stuck using very expensive interceptors and lacked a cheaper counter, then over the coming years the cost asymmetry would “really change the balance of power.”

What has happened so far (through late 2025):

• Houthi and other Iran‑backed forces have continued using relatively cheap drones and missiles to harass commercial shipping and U.S./allied naval assets in the Red Sea, forcing sustained air and missile defense operations by U.S. carrier strike groups and destroyers. Naval News’ detailed account of early Red Sea operations describes U.S. destroyers and carrier air wings expending large numbers of missiles and other munitions to defend against Houthi drones and anti‑ship ballistic missiles, validating the basic cost‑exchange problem Sacks described. (navalnews.com)
• Senior U.S. Navy leadership has publicly warned that the current heavy reliance on high‑end interceptors like SM‑3 and SM‑6 against such low‑cost threats is financially and logistically unsustainable in more intense or prolonged fights, again underscoring the cost asymmetry Sacks worried about. (businessinsider.com)
• Analysts estimate that defending against Houthi attacks has cost the U.S. several billion dollars, while the economic damage from the attacks themselves and the cost of the Houthis’ own munitions remain far lower, explicitly noting that "the cost‑exchange ratio of the campaign favors the Houthis" even though the conventional military balance still overwhelmingly favors the United States. (realclearworld.com)

However, the stronger claim that this cost asymmetry has already “really change[d] the balance of power” in favor of actors like the Houthis is not borne out:

• Expert assessments continue to emphasize that despite the favorable cost‑exchange ratio for the Houthis, the conventional military balance and broader regional balance of power still strongly favor the U.S. and its allies. The Houthis have become a serious asymmetric nuisance that can impose costs and disrupt shipping, but not a peer force capable of overturning U.S. naval dominance. (realclearworld.com)
• At the same time, the condition that “we don’t have a good response” is beginning to erode. The U.S. Navy has fielded and tested directed‑energy systems such as the HELIOS laser on USS Preble as a low‑cost counter to drones and potentially some missiles; only one destroyer is currently equipped, but it shows movement toward exactly the kind of cheaper defensive layer Sacks said was missing. (en.wikipedia.org)
• Other states are also investing heavily in similar low‑cost air and drone defenses (e.g., the U.K.’s DragonFire naval laser contract, Israel’s Iron Beam, and U.S. high‑power microwave systems like Epirus Leonidas), indicating a broader push to close the cost gap rather than passively accepting a permanent asymmetric advantage for cheap‑drone users. (reuters.com)

Netting this out: Sacks’ intermediate observation—that the Houthis are exploiting a dangerous cost asymmetry and that the U.S. has been using very expensive interceptors against cheap threats—has been validated. But his full prediction was about a substantial, longer‑term shift in the overall balance of power “over the coming years” if nothing changed. As of December 2025, the U.S. and allies still retain clear military superiority, and significant efforts are underway to develop cheaper counters. The time horizon he invoked has not fully elapsed, and the drastic balance‑of‑power shift he warned about has not clearly occurred.

Given that, the fairest assessment today is that the prediction’s ultimate outcome is still unresolved, so it is inconclusive (too early) rather than clearly right or wrong.

The prediction’s timeframe is “over the next decade” from the podcast date (April 12, 2024), i.e., roughly through 2034. As of now (December 1, 2025), only about 1.5 years have elapsed.

There is evidence that:

• Small attack and reconnaissance drones are proliferating and being used extensively in conflicts like Ukraine–Russia and in the Middle East, including loitering munitions and FPV drones.
• A variety of automated and semi‑automated counter‑UAS systems using sensors and computer vision exist, including gun‑based and turret‑like solutions (e.g., C-UAS remote weapon stations, vehicle‑mounted systems, and fixed-site defenses), and some are being trialed or fielded at bases and on vehicles.

However, the specific claim is that automated gun‑turret systems using computer vision to detect and shoot down drones—like the Bullfrog concept—will become a common and practical method deployed on vehicles and around installations. Whether such systems become widely common and standardized across militaries, critical infrastructure, and VIP protection by 2034 cannot yet be assessed in 2025.

Because:

• The end of the stated time horizon (2034) has not been reached.
• Current adoption levels and doctrine are still evolving rapidly, and it is too early to say whether these systems will end up as a niche, transitional, or truly ubiquitous solution.

the correct assessment at this time is that the prediction’s truth value is too early to call, hence inconclusive rather than right or wrong.

As of December 1, 2025, there is strong evidence that:

1. U.S. military recruiting shortfalls are real and persistent.

   • The U.S. Army, Navy, and Air Force all reported significant active-duty recruiting shortfalls in FY2023 and continued challenges into FY2024–2025, widely described by Pentagon officials and major outlets as a “recruiting crisis.”

2. The Department of Defense is explicitly pushing increased automation and large-scale use of unmanned systems.

   • In August 2023, DoD announced the Replicator initiative, aiming to field thousands of “attritable autonomous systems” across multiple domains (air, sea, land) within 18–24 months to counter China’s mass and change force structure at scale.
   • Service branches are accelerating procurement and experimentation with drones and other unmanned platforms (e.g., loitering munitions, unmanned surface vessels, autonomous ISR platforms), with senior leaders repeatedly citing manpower constraints and the need to offset adversary numbers as motivations.

However, Chamath’s prediction has two stronger elements that cannot yet be confirmed or falsified:

• Time horizon – He said “over the coming years,” which implies a multi‑year structural shift, not something that would necessarily be observable by late 2025.
• Degree of dependence – “Automate and become drone dependent” suggests that unmanned systems will become central to core military functions, not just important or rapidly growing. While the trajectory clearly points toward much heavier use of drones and automation, current U.S. force structure in 2025 is still predominantly manned and the system is in transition.

So far, the evidence supports the direction of his reasoning (recruiting crisis → push for automation and drones), but it is too early to say that U.S. forces have in fact become “drone dependent” for core functions, or that this end-state will definitively occur. The prediction is therefore not yet testable in a decisive way.

Given the short time elapsed and the inherently long-term nature of the claim, the fairest assessment as of December 1, 2025 is:

Result: inconclusive (too early to judge).