George Filippakis

Lawyer building with code

[George Filippakis @GeoFilippakis]

@0xartifax @harvey @FireworksAI_HQ Strong point. The model routing worked, but the "sound less like AI" prompt failed ¯\_(ツ)_/¯

[George Filippakis @GeoFilippakis]

@shl AI doesn't have anxiety.. it hallucinates confidence, which is arguably worse.

[George Filippakis @GeoFilippakis]

@SellersCounsel Nothing says "efficiency gain" like paying counsel to fact-check the robot's misunderstanding of MS Word markup.

[George Filippakis @GeoFilippakis]

The SLM vs LLM debate is especially relevant in legal AI because law firms don't just need intelligence. They need defensibility.

For some workflows, a smaller, task-specific model that is cheaper, easily auditable, and highly consistent may be more useful than a frontier model with broader capability but higher cost and governance complexity.

[George Filippakis @GeoFilippakis]

@rohitdotmittal Small ARR + serious logos = distribution arbitrage.

[George Filippakis @GeoFilippakis]

@SMB_Attorney It means the rainmakers are not the part of BigLaw AI is coming for first.

[George Filippakis @GeoFilippakis]

@gabepereyra @winstonweinberg That makes sense. Do you think model routing becomes an auditable governance decision in itself? Not just which model touched the matter, but e.g. why model A was permitted/selected for operation Y on matter Z, instead of another available model?

[George Filippakis @GeoFilippakis]

@boardyai Turns out changing the world requires a shocking amount of GPUs and optimism.

[George Filippakis @GeoFilippakis]

@milesdeutscher The AI bubble debate gets messy because "AI" includes both Nvidia printing money and frontier labs burning money (for now). Same narrative. Very different income statements.

[George Filippakis @GeoFilippakis]

@emollick The bottleneck has moved from model capability to organizational deployment.

[George Filippakis @GeoFilippakis]

Anthropic filing a draft S-1 is significant beyond the valuation story. A public Anthropic means audited financials, disclosed material risks, public litigation exposure, and regulatory scrutiny at a new level. The legal infrastructure around frontier AI is about to get a lot more interesting.

[George Filippakis @GeoFilippakis]

@gregisenberg Build anyway is probably the only advice with decent backwards compatibility.

[George Filippakis @GeoFilippakis]

@unusual_whales Claude is about to discover the most dangerous prompt of all: shareholder expectations.

[George Filippakis @GeoFilippakis]

@DanLinna We have officially reached the "objection, prompt injection" phase of legal AI.

[George Filippakis @GeoFilippakis]

@scottastevenson AI writing is bad when it's lazy. AI point of view is dangerous when it replaces having one.

[George Filippakis @GeoFilippakis]

@chiajy2000 @gooby_esq @sethtjf @sahiladhawade The firm reviewed the matter internally and concluded that no dispute exists.

[George Filippakis @GeoFilippakis]

@gooby_esq @sethtjf @sahiladhawade The final boss of legal services is a single global firm suing itself for fees.

[George Filippakis @GeoFilippakis]

@EMostaque Explains why some people hallucinate confidence and refuse feedback.