30초 요약
Berman shares a January 2024 video in which Mark Zuckerberg says Meta is training Llama 3 and combining AI research efforts. Zuckerberg describes ambitious future compute targets and a goal of making general intelligence widely available; Berman responds to those plans and miscalculates an illustrative hardware total.
- 1Research and general intelligenceZuckerberg describes bringing Meta’s AI research efforts closer together.
- 2Compute targetHe describes a planned 350,000 H100 GPUs and roughly 600,000 H100-equivalent compute by year-end.
- 3Llama 3Berman highlights Zuckerberg’s statement that the model was then in training.
- 4Hardware arithmeticThe host’s stated per-chip assumption yields $10.5 billion, before infrastructure costs.
핵심 정리
- Zuckerberg said in January 2024 that Llama 3 was being trained.
- He described responsible openness and useful AI as long-term goals, not a completed AGI release.
- His compute figures were targets for the end of 2024.
- Berman’s illustrative multiplication of 350,000 GPUs at $30,000 each equals $10.5 billion, not $1 billion.
챕터별로 다듬은 전문 읽기
Zuckerberg’s announcement
Zuckerberg discusses Meta’s general-intelligence goals, compute plans and Llama 3 training. He also describes glasses as a possible interface for AI.
Matthew Berman introduces a January 2024 announcement from Meta CEO Mark Zuckerberg, who says Llama 3 is in training. The clip records plans and goals at that moment, rather than a completed general-intelligence system.
Zuckerberg says Meta is bringing its two AI research efforts closer together to support a long-term goal of building general intelligence, making it broadly useful and releasing it responsibly. He identifies reasoning, planning, coding and memory among the capabilities that future assistants and other services would require.
Zuckerberg describes a compute target for the end of the year: around 350,000 Nvidia H100 GPUs, or approximately 600,000 H100-equivalent units of compute including other GPUs. He says Llama 3 is being trained and that Meta has a roadmap of further models; these are company plans as stated in January 2024.
He also sees AI and the metaverse coming together through new devices. In his view, glasses can let an assistant see and hear what the wearer does throughout the day, and he points to the early reception of Ray-Ban Meta glasses.
Berman’s reaction
Berman discusses the significance of the announcement and Meta’s compute investment. His arithmetic for the GPU example requires correction.
Berman highlights the direct confirmation that Llama 3 is in training and recalls the impact of earlier Llama releases on access to models. He views Meta’s statement as evidence of an intensified AI effort alongside its metaverse work. Availability of model weights, commercial permissions and full openness are distinct questions that depend on the relevant license.
Berman emphasizes the scale of Meta’s planned infrastructure and uses an assumed $30,000 per H100 to illustrate cost. His stated total of about $1 billion is an arithmetic error: 350,000 times $30,000 is $10.5 billion, excluding power and cooling. The calculation is illustrative, not evidence of Meta’s actual purchasing prices or expenditure.
Berman concludes that Zuckerberg sees general intelligence as an important goal and hopes Meta can build and openly release it. That is an ambition described in the announcement, not evidence that AGI had been achieved.
같은 채널의 다른 글
34:21EN
Matthew Berman· 2026년 10월 27일Matthew Berman Examines Anthropic’s IPO Filing, AI Growth and GovernanceMatthew Berman, Sam Altman · Forward FutureThe presenter reviews the S-1’s financial metrics, compute strategy and voting structure while questioning the tension between rapid AI development and safety.
0:26ENMatthew Berman· 2026년 10월 25일Sam Altman on Human Values and Problem-Solving in an AI WorldSam Altman · OpenAIIn a short interview excerpt, Altman considers the enjoyment of solving hard problems and the human role in interpreting results.
0:36ENMatthew Berman· 2026년 10월 25일Kimi K3.1 and a Rubik’s Cube Solver DemonstrationMatthew Berman · Forward FutureMatthew Berman shows a scrambled virtual cube being solved and gives the Kimi K3.1 test a favorable assessment.
0:21ENMatthew Berman· 2026년 10월 24일An AI-Generated Kimi Benchmark Chart Used Invented ScoresMatthew Berman · Forward FutureMatthew Berman asks for a graph of Kimi K3.1 Pro benchmarks, then discovers that the displayed figures are illustrative placeholders.
25:50ENMatthew Berman· 2026년 10월 24일Kimi K3.1 Pro Tested: Strong 3D Demos, Uneven Results and Competitive PricingMatthew Berman · Forward FutureMatthew Berman tests the open-weight model on coding, 3D rendering and visual explanations, then compares benchmark results and prices.
32:20ENMatthew Berman· 2026년 10월 23일Who Is Local AI For? The Costs, Trade-offs and a Hybrid FutureMatthew Berman · Forward FutureMatthew Berman weighs data control and independence against hardware cost, setup complexity and inference speed.