Victor Algaze

@burkeholland I’ve waited 8 years for this day… github.com/valgaze/redire… medium.com/hackernoon/uti…

I'm learning so much from watching agents do all my work. Like what the heck does 2>&1 do and why is it on every single terminal command.

@zirkelc_ @nicoalbanese10 @aisdk That’s cool (+ thorough!) The “dream-state” could be something that lives here and throws an InvalidArgumentError: github.com/vercel/ai/blob… Perhaps opt-out rather than opt-in

@valgaze @nicoalbanese10 @aisdk Interesting, thanks for sharing! I did the same at type-level, because it's not more flexible with different schema libraries. x.com/zirkelc_/statu…

OpenAI structured outputs don't support optional properties It took me a while to figure out why I was getting schema mismatch errors from @aisdk despite the schema being super simple If you're using Zod, you need to use `nullable()` instead of `optional()` The important difference is that the resulting JSON schema contains the `required` key which OpenAI needs

@nicoalbanese10 @zirkelc_ @aisdk The OpenAI Node SDK throws an error, mybe aisdk should do something similar for clearer errors? 🔗 #L48-L60" target="_blank" rel="nofollow noopener">github.com/openai/openai-…

@zirkelc_ @aisdk sadly I don't!

First off, what's TAU Bench? It's a clever benchmark for LLM agents in customer service domains, where the agent has to help a customer solve their problems (lost credit card, missed flights etc). Solving these problems involves reading from a database, making function-calls, and generally being able to communicate coherently with the customer. The novel part of this benchmark is that the customer is also an LLM! A funny quirk of this setup is that since most LLMs are trained to be assistants, the customer LLM sometimes reverts to its ground state and ends up helping the service agent 😅

@slow_developer Claimed vs effective ctx window length: github.com/NVIDIA/RULER arxiv.org/abs/2404.06654

"1m context window" models after 200k tokens

@the_ross_man @mattpocockuk @DSPyOSS @dosco @lateinteraction Ex:

@the_ross_man @mattpocockuk @DSPyOSS You might be thinking of @dosco github.com/ax-llm/ax Semi-endorsed by DSPy author @lateinteraction

Every time I post about prompting, someone in the comments mentions @DSPyOSS. I 100% cannot understand the value proposition If you have evals that test your system, why run complex RL optimizations that obscure your prompt? Please, change my mind

@colinhacks This ticks all your boxes: vitepress.dev/guide/what-is-…

I want a tool that accepts a single README dot md and generates a pretty one-page docs site. maybe a config file somewhere for OG metadata, etc. constraints: - support for GFM markdown - syntax highlighting - table of contents - light/dark mode - the README should still be pretty on GitHub (no mdx, no markdoc) does this exist?

@warmwaffle x.com/mitchellh/stat…

@thdxr @rodpadev I'm fairly certain that @mitchellh would understand this pain.

anthropic is at risk of making a big mistake it's something we've seen too many times before imagine having the crazy goal of building a platform - something thousands of companies and products are built on top of you realize just building the platform isn't enough, so you start to build tools that make it easier to use the platform and demonstrate its capabilities these tools get their own names and identities and teams working on them and very quickly these teams forget they only exist to drive people onto the platform and then one day someone external makes a tool that accomplishes that goal and does it even better it should be a moment of success - this is the original dream, to see great things built for your platform but structurally these teams have long forgotten that so it's a moment of competition. in the worst cases they even try to squash it we've experienced this building SST and how some teams at AWS saw our work as competitive even as we were driving dollars to AWS and tapping into a market they could never reach there are exceptions - cloudflare has invested resources in helping us even though they have wrangler, somehow their teams are setup in a way to not see us as a threat but it's a real test - we'll soon be able to see if anthropic as an org is really aligned with becoming a platform or if they fall into this same trap

@thebatdev @thdxr @SST_dev @DrizzleORM @nextjs @polar_sh @bunjavascript @expo Interesting pattern for VPC 👀👀 github.com/nitishxyz/stac…

Okay here it is, the SST monorepo with the following: stackforge.xyz infra: @SST_dev cloud: aws database: RDS Postgres ORM: @DrizzleORM (what else) auth: openauth frontend: @nextjs payments: @polar_sh everything powered by: @bunjavascript TBD: mobile: @expo (somewhat done) webhooks: (maybe using the api or lambda) telegram integration: for notifications maybe email: for auth codes. This template is for SST infra deployment primarily. eager to know if anyone would like to see a full sass app in this same monorepo.

@nmtmbr @CChristineFair Suspect @CChristineFair is referencing ~Feb 2019 incident: time.com/5564980/india-… Off ramp for both sides— f16 shoot down story was a victory, later return of downed pilot diplomatic gesture This was the pilot: en.wikipedia.org/wiki/Abhinanda…

@vegasstarfish Saw this mounted in the elevator lobby of the Signature towers last nite…

🚨 Las Vegas Shooting Situation at Signature Towers at MGM Grand. According to source, suspect is in custody.

@lihao20230505 @zswitten @AnthropicAI @lihao20230505 Is this prompt deployed as an agent anywhere?

@zswitten @AnthropicAI Do you think prompts like the following format useful?

The Prompt Doctor is in! I joined @AnthropicAI as a Prompt Engineer* a few weeks ago. To celebrate, I’m gonna do an AMA this Thursday from 10am-12pm PST to answer any and all your prompting questions. *And Librarian! Loved telling my in-laws I quit my job to become a librarian

@genmon @OpenAI There’s an aircraft carrier worth of compute power + astounding capabilities, but often the last few “feet” turn into unstructured text parsing & wrangling JSON

If anyone from @OpenAI is listening, then what would be insanely useful would be `gpt-3.5-turbo` but instruction-tuned to always reply in JSON according to a Typescript interface in an opening system message 🙏

@voiceflow @Cisco Stachin’ up a storm lol

🆘 Engineers can't stand static conversational AI artifacts. Take it from @valgaze - a CX software engineer at @Cisco.

On Thursday night @GrantDKeller and I showed off the latest version of our software at @Expert_Dojo put on by @AILA_Community Video by @valgaze We are looking forward to showing it at more events #interactive #ai #aiart

@OfficialLoganK @OpenAI 1. - (Per @AndrewMayne) collection of fun shortcut keywords like “tags” or “tl:dr;” that can reduce token usage 2. Guidance on “ special” syntax or characters that can influence the system output Ex. Prompt’ing on other systems:

Working on an updated prompt engineering 🎙️ guide for the @OpenAI docs, send your best resources or suggestions to be potentially included 🧵👇

@BowTiedFun @danielgross Was about to say how “lovely” the current strategic position vs competitors looked but cut off

@danielgross What is a lovel?

@nonmayorpete @bakztfuture @EnglishOER

Master's program. 2 grants to study it. 30K in debt if you get kicked out. Yeah, use ChatGPT.