$SIRI Sirius XM Holdings CEO: YouTube expands SiriusXM’s addressable advertising audience to 255M monthly listeners, equivalent to ~90% of Americans aged 13+ “With YouTube now, we're going to be at 255 million monthly listeners, and that's 90% of the U.S. population above the age of 13. So it's just -- it's the scale of the audience that we have, the scale of the types of content we have and the scale of the advertiser relationships that I think maybe are not fully appreciated.”

Graphs reveal area secrets through simple shapes. The constant function f(x)=a spanning 0 to xₘ encloses a rectangle of area a𝑥ₘ. The linear function f(x)=a𝑥 encloses a triangle of area a𝑥ₘ²/2. These equal the evaluated definite integrals of each. Spring manufacturers compute elastic potential energy as the triangular area under the force-extension graph, yielding ½kx² for a spring constant k.

this touches a deeper aspect of current AI research: if the same core theory is applied to study different questions, does that count as plagiarism? I'd probably say no, especially in this case. The XM paper cites IMLE right before introducing its core objective (Eq. 1), and Appendix E.3 argues that IMLE is a specific instance of end-to-end Forward XM. It also pushes back on IMLE's theory, arguing the working mechanism was never implicit maximum likelihood but the multi-candidate search itself. In this case I think novelty (or contribution) lives more in the question, not just the method. IMLE asked how to avoid mode collapse in conditional image synthesis. XM asks whether the same best-of-K objective (sample K candidates, backprop only through the one closest to the data) works as a third pre-training axis, with gains that grow with scale rather than saturate. The IMLE line of work never pursued these questions. Probably @AlexiGlad should have called out IMLE more in Section 3 and said plainly that Eq. 1 is the conditional IMLE objective (which I personally would also find it a bit odd). But this does not make it plagiarism. If reusing a core mechanism to answer new questions were plagiarism, much of modern ML would be guilty.

Practices for Integrating AI Agents into Enterprise Systems 【MCP Gateway / Tool Federation】 💡 Catchy Message "5 agents x 10 SaaS products = 50 custom integrations. This multiplication nightmare is what the MCP Gateway eliminates." Every new agent and every new SaaS connection compounds integration cost. Tool definition sprawl, schema inconsistencies, and silent API breaking changes -- these problems demand an architectural solution. 🔥 Problems Solved - N (agents) x M (SaaS) integration cost explosion - Duplicate and inconsistent tool definitions across agents - Indirect prompt injection through tool I/O - Tool selection accuracy degradation when too many tools are exposed to an agent - Silent SaaS API changes (schema drift) causing agents to process incorrect data 🏗️ The Pattern Bundle each SaaS connector as an MCP (Model Context Protocol) server behind a gateway that manages tool discovery, authorization, call auditing, and scope control. Dynamically filter tool allow-lists by principal (department x agent type), exposing only the minimum necessary tools to each agent. Dangerous tools (delete, transfer funds, external send -- irreversible operations) get approval hooks. Tool definitions and API schemas are versioned as "contracts," periodically validated against live APIs to detect drift. Backward-incompatible drift triggers alerts and automatic tool deactivation as a fail-safe. ✅ When to Adopt - Use when: 10+ SaaS integrations. Multiple agents share common tools. Struggling with N x M integration complexity. - Skip when: Single-purpose agent with 2-3 fixed tools (direct integration is simpler and more robust). APIs are stable with extremely low change frequency. ⚠️ Pitfalls - Exposing 20-30+ tools to a single agent degrades tool selection accuracy. Use tool RAG for dynamic filtering or split into role-specific sub-agents. - Without contract testing (drift detection), you won't notice SaaS API changes until agents silently process incorrect data. Salesforce field changes happen more often than you think. - Deferring MCP server authorization design leaves all agents with access to all tools -- an open invitation for misuse. 🛠️ Implementation Approach - Build MCP servers for each SaaS (Salesforce, ServiceNow, Jira, Slack, Box, etc.). Adopt official MCP servers where available; otherwise auto-generate tools from OpenAPI specs and wrap them as custom MCP servers. - Deploy an MCP gateway with a tool registry (catalog). Index all tools from each MCP server and configure allow-lists filtered dynamically by department x agent type. - Set up OAuth 2.1-based authorization with approval hooks. Attach approval gates (linked to P09 dynamic authorization PDP) to dangerous tools (delete, fund transfer, external send) so they never execute without human approval. - Build a drift detection pipeline using contract testing (Pact, etc.) and a schema registry. Run weekly reconciliation between tool definitions and live API schemas; auto-deactivate tools and alert on backward-incompatible changes. - Control per-agent tool exposure to under 20 using tool RAG or role-specific sub-agent splitting. Dynamically filter tools by intent to maintain selection accuracy. #AIAgents# #EnterpriseArchitecture#

The bandwidth latency tradeoff: Solana blocks are split into FEC sets, each of these sets is broadcast through rotor/turbine. Turbine splits FEC set into 32 pieces and then adds 32 more shreds containing erasure coding to form 64 shreds then sends each of these shreds out to the validator set through a specific randomly chosen turbine path. How long does it take for each FEC set to get from the leader to a specific validator over turbine? Because of erasure coding, the validator does not need every shred. It only needs enough distinct shreds to decode. In the simple 32-of-64 case, the leader sends 64 shreds, but a validator only needs any 32 of them. So the slowest 32 paths do not matter for decoding. We can model this mathematically: for validator i, define Lᵢ as the latency distribution induced by: leader → random root → validator i where the root is sampled according to stake. (this is technically rotor not turbine but its just a simplification, you can do the same trick for turbine but the equations are messier). Each shred samples one relay path from Lᵢ. So in the 32-of-64 case, validator i observes X₁,…,X₆₄ ∼ Lᵢ These are the arrival times of the 64 shreds. But the relevant arrival time is Bᵢ = X₍₃₂₎ the 32nd order statistic or the time when the 32nd fastest shred arrived, completing the FEC set. We can write the cumulative distribution of X₍₃₂₎ as: Pr[Bᵢ ≤ t] = ∑ⱼ₌₃₂⁶⁴ (64 choose j) Lᵢ(t)ʲ(1−Lᵢ(t))⁶⁴⁻ʲ More generally, if a slice has m data shreds and p coding shreds, then validator i sees X₁,…,Xₘ₊ₚ ∼ Lᵢ and can decode once m have arrived Bᵢ = X₍ₘ₎. This turns Turbine design into a quantile/bandwidth tradeoff. Let n = m+p and m/n → q. Then Bᵢ ≈ Lᵢ⁻¹(q) and by the CLT for order statistics, √n · (X₍ₘ₎ − Lᵢ⁻¹(q)) ⇒ N(0, q(1−q)/fᵢ(Lᵢ⁻¹(q))²) TLDR More coding shreds -> block arrives faster! With more large validators moving to larger NICs and XDP activated, should we crank up the fan out to make the leader handoff faster if it costs us some theoretical max throughput?