ted hall

Dez caught it.

[New England Patriots @Patriots]

WR1

[ted hall @tedhall23]

@DSPyOSS Totally agree! DSPy way ahead of its time.

In the world of agentic workflows, how do you see the balance of more deterministic AI workflows vs. fully agentic, tool calling behavior, playing out for actual companies? I.e. cost, latency, etc.

[DSPy @DSPyOSS]

everyone in 2022: prompt tricks!!
us: D-S-P

2023: prompt engineering!!
us: DSPy

2024: agentic workflows!
us: DSPy

2025: context engineering!!
us: DSPy

2026: harness engineering!!!
us: .. DSPy

2027: some other ephemeral BS tinkering
…

[Football @BostonConnr]

Watched this play no less than 50 times

Charlie McAvoy — dawg.

[ted hall @tedhall23]

@mattadler81 @CatalystHotdogs

[Matt Perrault @sportstalkmatt]

Dime. What a shot

[Julian Edelman @Edelman11]

New England vs Everybody #GoPats

[ted hall @tedhall23]

@ChenMoneyQ @lateinteraction @isaacbmiller1 @tomehirata @tedcruz FWIW still running into the NoneType error when using gpt-oss-20b in the same codebase. @ChenMoneyQ

[ted hall @tedhall23]

@ChenMoneyQ @lateinteraction @isaacbmiller1 @tomehirata @tedcruz @lateinteraction @ChenMoneyQ Thanks! I got this working with GPT-5.1 and 5.2.

I put together a DBX demo showing that with large contexts dspy.predict consistently hits TPM rate limits, while dspy.RLM does not — even w/ same DSPy Signature + Model

https://t.co/nhK9mpjKZ7

[ted hall @tedhall23]

@lateinteraction @isaacbmiller1 databricks-gpt-oss-20b

[ted hall @tedhall23]

Tony Skinn and boys do it again taking down VCU 86-80 in EBA (16-1; 4-0 A10)

Final Four appearance 20 years later?
#PatriotPressure

[ted hall @tedhall23]

@klutchinggg

[ted hall @tedhall23]

Bama is rattled.

Kirby coaching circles around DeBoer.

Go Dawgs.

[ted hall @tedhall23]

I think VJ Edgecombe might be good at basketball.

[alex zhang @a1zhang]

What if scaling the context windows of frontier LLMs is much easier than it sounds?

We’re excited to share our work on Recursive Language Models (RLMs). A new inference strategy where LLMs can decompose and recursively interact with input prompts of seemingly unbounded length, as a REPL environment.

On the OOLONG benchmark, RLMs with GPT-5-mini outperforms GPT-5 by over 110% gains (more than double!) on 132k-token sequences and is cheaper to query on average.

On the BrowseComp-Plus benchmark, RLMs with GPT-5 can take in 10M+ tokens as their “prompt” and answer highly compositional queries without degradation and even better than explicit indexing/retrieval.

We link our blogpost, (still very early!) experiments, and discussion below.

[ted hall @tedhall23]

Memphis’ Sutton Smith is the next James Cook.

NFL Talent

[ted hall @tedhall23]

@MstreetAlmanac Let’s go!

[Yann LeCun @ylecun]

Too often, we think a task is easy because some animal can do it.
But the reality is that the task is fiendishly complex and the animal is much smarter than we think.
Conversely, we think tasks like playing chess, calculating an integral, or producing grammatically correct text are complex because only some humans can do them after years of training.
But it turns out these things aren't that complicated and computers can do them much better than us.

This is why the phrase "Artificial General Intelligence" to designate human-level intelligence makes absolutely no sense.