Twitxy
Saved
@Tyler_Marques's profile picture
View on X

Tyler Marques

@Tyler_Marques

Data Scientist / Engineer, Techie, Home-labber
Toronto, Canada tylermarques.com Joined May 2008
72 Following 187 Followers 55 Posts

Latest posts 5

Tyler Marques @Tyler_Marques
More AI playing games - love to see and it love to support it.

Autoresearch is a huge part of how we can push the frontier, and @alxai_ talks about how we apply it in our post here: goodstartlabs.com/research/keep-…
goodstartlabs.com Keep going: GPT-6 Astra vs Claude Opus 5.5, live for 48 hours GPT-6 Astra in Codex and Claude Opus 5.5 in Claude Code each write a StarCraft bot and climb a ladder of human-written bots, live for 48 hours.
alex duffy Verified account @alxai_ Oct 2
More & more AI is given a goal
& told to keep going

Benchmarking AI on this auto-research,
understanding tendencies & strategies,
directing them towards our goals,
is very important.

Today we're kicking off the Star Skirmish experiment
Opus vs Astra, write c++, win Starcraft

For 15 years humans have been writing algorithms that can manage whole Starcraft Broodwar army to compete against one another..

Lets see how well Astra (in codex) and Opus 5.5 (in Claude code) do against each other, and against the human records!
Video preview
View quoted post on X
Tyler Marques @Tyler_Marques
Excited for our friends over at @typesafeai for their public launch of Jev!

Efficiency is our biggest use case - we can do the same work for 0.05% the cost. See how we use Jev here: goodstartlabs.com/research/verif…
goodstartlabs.com Verification is the bottleneck | Good Start Labs As models do more, we need better ways to see what they’re doing and what we’re teaching them.
Diogo Almeida Verified account @CompleteSkeptic Sep 15
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?

I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev

• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions

AFAICT the shortest path to AI-based economic revolution
Video preview
View quoted post on X
2 replies 0 reposts 11 likes 1.2K views View on X
Tyler Marques @Tyler_Marques
@OpenAI's GPT-6 Astra sits at the top of our humor leaderboard, just barely edging out Fable 5.1. Both are just tipping over the 50% line.

They've got a lot of tools, but sense of humor still isn't one of them.
alex duffy Verified account @alxai_ Sep 9
Two years of AI progress. Still struggling to get the joke.

On 14,000+ real @playbadcards hands, Astra tops our updated LOL-Alignment benchmark
matching the human judge 50% of the time

Just 2% ahead of GPT-4o after 2 years!

Humor is subjective. Games make it measurable.

Funnier than AI?
Our team got ~60% on the examples.
Try them 👇
goodstartlabs.com/benchmarks/lol…
Video preview
View quoted post on X
Tyler Marques @Tyler_Marques
We launched twitch.tv/ai_diplomacy today! Been working on this for a while and super proud of it. Watch Claude, Gemini, o3, and others battle it out in the classic board game of Diplomacy.

Super proud to be working with @alxai_ and the team at @every
www.twitch.tv AI_Diplomacy - Twitch LLMs playing the classic game of Diplomacy. Watch as they converse, conspire, and betray each other. Find more info here: https://every.to/diplomacy Help us pay for more games by donating on Patreon!...