Tag

Tom's Hardware

3 issues found

Sep 18, 2026

Memory Gates Agents, Capital Funds Them

Description

Memory Gates Everything Chroma's 18-model eval found "context rot" degrading accuracy on trivial tasks; HuggingFace and IBM frame recall as the real limit.
Capital Meets Compute Mistral's €3B Series D — Europe's largest equity round — funds data centers and sovereign inference, not new model capability.
Typed Decisions Spread Jev's claimed 20-200x speedups (one independent test: ~25x faster, 580x cheaper) are landing in agent stacks via MCP bridges.

The Harness Eats the Model

Description

The Bottleneck Moved — Across every source, one truth dominates: raw model capability is no longer the constraint. OpenAI's Jalapeño chip undercuts Nvidia's flagship at a fraction of the power draw, Apple's M5 Ultra clusters hit 4.8TB/s aggregate bandwidth on a desk, and Qwen is teasing sparse architectures with just 6B active parameters. The question isn't "what model?" anymore — it's "what harness, what hardware, what control plane?"
Harness Is the New Frontier — SWE-bench Pro data shows swapping harnesses moves pass@1 from 23% to 52% on the same model. IBM's DABStep finds SOTA agents at just 14.55% on hard data tasks, while Shopify's CEO threatens to ban Claude over AGENTS.md failures. Instruction fidelity, cost control, and reliability — not raw capability — are the binding constraints.
Open-Weight Acceleration — DeepSeek's V4-Pro and V4-Flash bring 1M-token native context with a price-performance swing that "alters everything we knew," and Qwen's sparse n-gram tables could make frontier-ish capability genuinely local. But broken docs, mixed NIST evals, and weak agentic benchmarks temper the hype.
Eval Layer Is Catching Up — A wave of honest benchmarks (ScarfBench's sub-10% on enterprise migrations, ScreenSuite's 13 unified tests, Holotron-12B jumping from 35.1% to 80.5% on WebVoyager) is finally separating real capability from demo-day optimism. The next round of agent gains will come from engineering memory, harness, and eval layers — not bigger models.
Agents Training Agents — SF Compute's CEO cuts to the core: "You're gonna get the models themselves that will train the models." With coding agents producing training data and local inference making private loops viable, the human bottleneck shifts from research skill to orchestration. Secure enough compute, or die.

AI's Search for a Business Model

Description

The AI gold rush is getting expensive. This week, the conversation shifted from a breathless pursuit of capabilities to a sobering look at the bottom line. On one side, you have giants like Cohere dropping Command R+, a powerful model aimed squarely at enterprise wallets, a move celebrated and scrutinized across the tech sphere. On the other, the open-source community is in the trenches. On HuggingFace, developers are feverishly fine-tuning Meta's Llama 3 for every conceivable niche, while Reddit and Discord are filled with builders wrestling with the brutal realities of inference costs and vector database performance. The battle for the future of AI isn't just about who has the smartest model; it's about who can build a sustainable business. Nowhere is this clearer than the fierce debate around AI search, where startups are discovering that disrupting Google is more than just a technical challenge—it's an economic war. This is the moment where the hype meets the spreadsheet.