<pre><code> FROM hn.hacker_news
 LIMIT 100
</code></pre>
&quot;Oops I forgot the limit clause and now owe MotherDuck and OpenAI $93 billion.&quot;

I love the simplicity of this. Hurray for small models for small tasks.

SQL functions can be non-deterministic just fine. E.g. SQL:2003 grammar defines DETERMINISTIC | NOT DETERMINISTIC characteristic for CREATE FUNCTION. Or, e.g. PostgreSQL has IMMUTABLE | STABLE | VOLATILE clauses.

Not at all. Even the ones that provide a &quot;seed&quot; parameter don&#x27;t generally 100% guarantee you&#x27;ll get back the same result.My understanding is that this is mainly down to how floating point arithmetic works. Any performant LLM will be executing a whole bunch of floating point arithmetic in parallel (usually on a GPU) - and that means that the order in which those operations finish can very slightly affect the result.

They are not, necessarily. Especially when using commercial providers who may change models, finetunes, privacy layers, and all kinds of other non-foundational-model things without notice.

Aren&#x27;t LLM outputs deterministic given the same inputs?

Interesting -- is there any impact from LLM outputs not being deterministic?

The Prompt() Function: Use the Power of LLMs with SQL