November 9th • 3h 14m
Shownotes
Every new episode of No Agenda is accompanied by a comprehensive list of shownotes curated by Adam while preparing for the show. Clips played by the hosts during the show can also be found here.
J6
Gen Zed
Big Tech AI and the Socials
Running AI at home
Grok 4 is a proprietary frontier AI model from xAI with rumored specifications around 1.7 trillion parameters in a Mixture-of-Experts architecture, based on various reports. It achieves top-tier benchmark scores, such as 88% on GPQA Diamond and 61.9% on USAMO'25, placing it among the leading AI models for reasoning, coding, and general intelligence. However, its weights are not publicly available for local use, so you cannot run Grok 4 itself at home. Instead, to approximate similar AI performance (e.g., high-quality responses in reasoning, coding, and multimodal tasks), you'd rely on the largest open-source models with comparable capabilities, such as Meta's Llama 3.1 405B or similar-scale alternatives available by late 2025. These can be run locally but won't match Grok 4 exactly on all benchmarks or features like real-time web/X integration.
Achieving cloud-like performance at home involves trade-offs: local setups are limited by hardware constraints, power draw, heat, and inference speed (e.g., 5-20 tokens/second vs. near-instant cloud responses). You'd need a powerful PC or small cluster optimized for AI inference, using frameworks like Hugging Face Transformers, vLLM, or Ollama. Below is a breakdown of typical costs.
### Hardware Costs
Hardware is the main expense, as it must provide enough VRAM (GPU memory) to load and run large models. For a 405B-parameter model (a rough proxy for frontier-level performance), you'd need ~200-250 GB of total VRAM when using 4-bit quantization (a compression technique to reduce memory use without major quality loss). This requires multiple GPUs, as no single consumer GPU has that much VRAM. CPU-only setups (e.g., via distributed clusters) are possible but much slower, making them unsuitable for "performance" matching cloud speeds.
- **Key Factors Influencing Cost**: New vs. used GPUs (e.g., used A100/H100 on eBay can halve prices but risk reliability). Cooling, power supply upgrades, and storage (e.g., 2-4 TB SSD for model files) add $500-2,000. For MoE models like Grok approximations, active parameters per inference are lower, potentially reducing VRAM needs by 30-50%.
- **Not Feasible for Exact Match**: If Grok 4 is truly ~1.7T parameters, even quantized it'd require 800+ GB VRAM—impractical at home without a data-center-like setup costing $100,000+ (e.g., 10+ H100 GPUs). Most home users cap at 405B-scale for now.
### Software Costs
- **$0 (Free)**: Open-source models (e.g., Llama, Mistral, or xAI's older Grok-1 if applicable) are downloadable from Hugging Face. Frameworks and tools like Ollama, LM Studio, or PyTorch are free and open-source. No licensing fees, though optional paid APIs (e.g., for fine-tuning) could add $10-50/month if needed.
In summary, a realistic home setup for near-Grok performance starts at $10,000-$20,000 in hardware (mid-range rig) with free software, but expect compromises on speed and scale compared to xAI's cloud infrastructure. For most users, subscribing to Grok via x.com Premium+ ($16/month) or SuperGrok ($5-10/month, details at https://x.ai/grok) is far cheaper and easier than building locally. If aiming for exact replication, it's currently not viable at home without enterprise-level investment.
Shutdown
The filibuster rule
In the U.S. Senate, the filibuster is a procedural tactic that allows senators to extend debate indefinitely on certain matters, effectively delaying or blocking a final vote unless a supermajority ends the debate. This is governed by Senate Rule XXII. To overcome a filibuster, senators must invoke "cloture," which requires a three-fifths majority of all senators duly chosen and sworn—typically 60 votes out of 100—to limit further debate and proceed to a vote. This 60-vote threshold does not apply to the final passage of most items (which usually needs only a simple majority of 51), but rather to ending the debate that precedes it.
### Types of Senate Actions or Votes Subject to the Filibuster (Requiring 60 Votes for Cloture)
The filibuster can apply to any debatable matter, meaning these require 60 votes to invoke cloture and advance:
- **Legislation and Bills**: Most substantive bills, joint resolutions, and concurrent resolutions, including those on policy issues like healthcare, immigration, or appropriations (outside of budget reconciliation).
- **Amendments**: Proposed changes to bills or resolutions during floor debate, which can be filibustered if not germane or time-limited.
- **Motions to Proceed**: Requests to bring a bill or measure to the Senate floor for consideration, which are often debatable.
- **Conference Reports**: Final versions of bills reconciled between the House and Senate.
- **Other Motions and Measures**: This includes motions to reconsider, unfinished business, or even certain procedural questions like approving the Journal, if they become subject to extended debate.
In practice, this makes the 60-vote cloture a de facto requirement for advancing most non-exempt Senate business, as a minority can otherwise stall progress through "silent filibusters" (threats of delay without actual talking) or traditional talking filibusters.
### Exceptions: Actions Where the Filibuster Does Not Apply or Cloture Uses a Different Threshold
Certain procedures bypass the filibuster entirely through statutory time limits on debate, nondebatable motions, or Senate precedents, allowing advancement and passage with a simple majority (51 votes). Other cases have altered thresholds via the "nuclear option" (a parliamentary maneuver to change rules by majority vote). Over 160 such exemptions exist, created since 1969. Key exceptions include:
- **Nominations**:
- Executive branch positions (e.g., cabinet secretaries) and lower federal judicial nominations: Simple majority for cloture since a 2013 Democratic-led rule change.
- Supreme Court nominations: Simple majority since a 2017 Republican-led change.
- Post-cloture debate is limited (e.g., 2-30 hours depending on the role), but no 60-vote hurdle.
- **Budget Reconciliation**: Used for tax, spending, and debt-limit changes tied to the annual budget resolution. Debate is capped at 20 hours, with no filibuster allowed; passes by simple majority (as seen with the 2017 Tax Cuts and Jobs Act or 2022 Inflation Reduction Act). Subject to the Byrd Rule to prevent extraneous provisions.
- **Expedited Statutory Procedures** (debate time-limited, motions often nondebatable):
- **Congressional Review Act**: For disapproving federal regulations; 10-hour debate cap.
- **War Powers Resolution**: Joint resolutions on military engagements; 3-day consideration limit.
- **National Emergencies Act**: Resolutions terminating emergencies; automatic committee discharge and limited debate.
- **Trade Promotion Authority (Fast-Track)**: Implementing trade deals; 20-hour debate limit.
- **District of Columbia Home Rule Act**: Disapproving D.C. laws; 10-hour debate.
- **Arms Sales and Military Base Closures**: Specific fast-track rules with time caps.
- **Debt Limit Increases**: Certain resolutions with 10-hour debate limits.
- **Changes to Senate Rules**: Cloture here requires a two-thirds majority of senators present and voting (not fixed at 60).
- **Other Nondebatable Items**: Routine matters like motions to adjourn, table, or postpone indefinitely are decided by simple majority without filibuster risk.
These rules can evolve through Senate action, but as of November 2025, the filibuster remains intact for most legislation despite ongoing debates about reform.
Ending it
To end the filibuster rule in the U.S. Senate (specifically, to modify or eliminate Senate Rule XXII, which governs cloture and debate), there are two primary pathways, each with different vote thresholds:
### Standard Procedure (Cloture on Rule Changes)
- Under the current Senate rules, changing the standing rules requires invoking cloture on the proposed rule change.
- This cloture vote needs a **two-thirds majority of senators present and voting** (not a fixed number like 60, but typically around 67 if all 100 senators are present).
- After cloture, the actual vote to adopt the rule change would require only a simple majority (51 votes if all are present).
- This higher threshold makes it difficult to end the filibuster through regular order, as it effectively allows a minority to filibuster the change itself.
### Nuclear Option (Parliamentary Maneuver)
- Senators can use a procedural workaround called the "nuclear option" to bypass the two-thirds cloture requirement and change rules with a lower threshold.
- This involves raising a point of order (e.g., arguing that the filibuster doesn't apply to certain matters), and if the presiding officer rules against it, appealing that ruling.
- Overriding the presiding officer's ruling requires only a **simple majority** (51 votes if all senators are present and voting).
- This method has been used successfully in the past:
- In 2013 (Democrats) to eliminate the filibuster for most executive and judicial nominations.
- In 2017 (Republicans) to extend that to Supreme Court nominations.
- It could theoretically be applied to eliminate the legislative filibuster entirely, but as of November 2025, it hasn't been used for that purpose, and the filibuster for legislation remains in place despite reform discussions.
Ending the filibuster would require bipartisan support or a party with a strong majority willing to use the nuclear option, as it could set a precedent for future majorities to pass legislation more easily. Senate rules can only be changed at the start of a new Congress (every two years) or mid-session with agreement, but the thresholds above apply in either case.
