Jack Hogan

AI Research Scientist

[Jack Hogan @j_h0gan]

@johnowhitaker Best of luck Johno! You were one of the reasons I wanted to join answerai and I've no doubt you'll do something completely unique and interesting next

[Jack Hogan @j_h0gan]

@MLStreetTalk @zzznah @Esychology @johnowhitaker people are still working on NCA!

[Jack Hogan @j_h0gan]

@natolambert @allen_ai @interconnectsai *typos

[Jack Hogan @j_h0gan]

The "Build UIs" section of that post was inspired by you @HamelHusain! You always say to build UIs to make it easy for humans to inspect traces for evals and it inspired me to build UIs to make it easy for us to test our MCP tools

[Hamel Husain @HamelHusain]

@jeremyphoward @j_h0gan @htmx_org Wow its really fancy, even has a glass effect thingy

[htmx.org / CEO of htmx 4 (same thing) @htmx_org]

one file enthusiasts, rise up!

[Jack Hogan @j_h0gan]

@TommyYesItsMe Thank you, fixed!

[Jack Hogan @j_h0gan]

In 2023, I decided I wanted to build a personal website/blog. I knew nothing about web dev but felt emboldened by AI. Sadly, my excitement quickly waned trying to debug the thousands of LOC and understand the framework soup, so I parked the idea. 2 yrs later, I've had another go.

[Jack Hogan @j_h0gan]

@math_rachel Your post on "Making Peace with Personal Branding" convinced me to start my own blog. After years of procrastination, I launched it just yesterday, so thank you!
https://t.co/0ZPkMurk29

[Jack Hogan @j_h0gan]

I was one of the lucky few in last year's course and I've already signed up to return. Every lesson you'll learn something new about programming, problem solving, web dev, etc. But more importantly, you'll develop an approach to AI that keeps you learning instead of atrophying.

[Jack Hogan @j_h0gan]

99% of AI "guidance" online is of the (fantasy) genre "10x your productivity by running parallel Claude Codes that orchestrate sub-Claude-Codes that dynamically spin up sub-sub-Claude-Codes". @jeremyphoward's solveit is the opposite and I can't recommend his course highly enough.

[Jeremy Howard @jeremyphoward]

It's a strange time to be a programmer—easier than ever to get started, but easier to let AI steer you into frustration. We've got an antidote that we've been using ourselves with 1000 preview users for the last year: "solveit"

Now you can join us.🧵
https://t.co/GLKm0woI8b

[Jack Hogan @j_h0gan]

Most AI tools give the illusion of rapid progress but results that don't surive first contact with testing. The @answerdotai team have done an incredible job designing solveit in a way that makes it genuinely hard to fall into the traps and dead ends of most "vibe coding" tools.

[Jack Hogan @j_h0gan]

Have only used @AnthropicAI's Sonnet 4.5 a few times but a strange quirk I'm noticing is it tends to use language that suggests it's an actual developer

[Jack Hogan @j_h0gan]

@fchollet Unfortunately, shipping to the UK seems to be $59

[Jack Hogan @j_h0gan]

@adityaag This is why I'm so excited to be taking @TheZachMueller's "scratch to scale" course—bringing all the tricks of the trade out into the open!

[Jack Hogan @j_h0gan]

That may not seem like much but as we increasingly feed entire codebases into LLMs, concise code makes a huge impact. I love that while docments might seem like a response to token limits, it in fact predates LLMs — humans shouldn't waste time writing things twice either!

[Jack Hogan @j_h0gan]

A nice feature of OpenAI's new Agents SDK is the automatic parsing of function signatures into a schema. Before now, it was very annoying having to separately define both... Unless you were one of the people who realised that @jeremyphoward already solved this 11 months ago

[Jack Hogan @j_h0gan]

Reading and writing functions in this way is annoying as a human — and for LLMs it's just wasted tokens repeating the same information twice. numpy-style above is 79 GPT-4o tokens; docments-style is only 57.