Litigation attorney slams Anthropic for opaque Claude model differences in legal work
addicted2soysauce · reddit · 2026-09-18
An attorney using Claude for document analysis and legal writing criticizes Anthropic's opacity. Key points:
- Legal work needs predictable answers (does a contract term exist? is breach evidence in 2,000 pages?), but a 30-minute Claude run ending in "I don't know" is ambiguous: no evidence, misunderstood request, or missed evidence.
- Analyzing survey drawings requires inferring how they were drafted — beyond simple visual overlay.
- Unexplained model differences (e.g., Opus returning 12 court opinions vs. another model's 14) can affect multi-million-dollar litigation, and verifying output means re-doing the analysis.
Proposals: shallow explanations of model behavioral differences, an NDA-gated "circle of trust" tier with security audits akin to LexisNexis background-check software, or a premium tier with leak-bond reimbursement. He notes Anthropic's costs are trivial versus his fees and would pay for determinism.
More from Models
- Jev claims first model optimized for low/medium-entropy tasks; frontier models only for high-entropy — multiply_matrix · 2026-09-18
- Image generation used Transformers for years; autoregression plus language-only pretraining is the novelty — benno_krojer · 2026-09-18
- How TypeSafe Jev turns LLM token probabilities into snap judgments — and why frontier labs could clone it — JnBrymn · 2026-09-18
- Cryptographer mocks OpenAI: sandboxes can't even run a recent Linux kernel — matthew_d_green · 2026-09-18
- GPT-6 weekly limits spawn 'usage maxing': users race the reset timer — rschu · 2026-09-18
- Researcher: real-time driving-game models were doable in 2016 — industry has LLM tunnel vision — gandamu_ml · 2026-09-18