Friea Berg

Head of Growth & Strategy, Snorkel AI. Opinions mine.

[vincent sunn chen @vincentsunnchen]

https://t.co/aNuEf6Yu9j

[Alex Shaw @alexgshaw]

Thanks @SnorkelAI for the great tasks and especially @fredsala, Tom Walshe, and Jeong Shin for the collaboration

Terminal-Bench 2.0 on the horizon 👀 + some other exciting releases!

[Friea Berg @friea]

@kevinweil @VarshineSri @tejalpatwardhan 👏 @tejalpatwardhan

[Friea Berg @friea]

@phoebethacker Congratulations to the entire team - love the diversity of occupations! Looking forward to insights on how different models perform on practical every day tasks.

[Friea Berg @friea]

@embirico Love this!

[Friea Berg @friea]

@xai @radhakrishnanit 🚀🚀🚀

[Percy Liang @percyliang]

-2016 (classic era): focus on data efficiency
2017-2025 (pretraining era): focus on compute efficiency
2026-: focus on data efficiency (again)
The standard Transformer paradigm is optimized for compute efficiency. As we look at data efficiency, we'll see very different design decisions, which will be exciting!

[Friea Berg @friea]

Are you exploring building RL environments for agentic use cases? Here's a good breakdown of misconceptions to avoid surprises.

[Alex Ratner @ajratner]

Lots of chatter about agentic/RL simulation environments recently!

Some key misconceptions (slightly caricatured):

>> Building RL envs is easy, because you just code up a verifier quickly, and let the model do the tough data generation on its own!
- Usually, this boils down to over-indexing on environments where verification is easy.
- For example: you might need a chess expert to generate realistic expert gameplay traces, but anyone with a basic chess rulebook could verify a win easily.
- However: there are many, many settings where verification is not at all trivial. The simplest examples are settings with nuanced, domain-specific evaluation rubrics (e.g. most real world enterprise settings). An extreme example being: verify whether a program will halt :)

>> Building RL envs will get commoditized as the "standard" environments get rapidly solved.
- RL environments effectively encode a complete product spec - including unique tools, data resources, constraints, rubrics/verifiers, and human/agent simulators - and as such, are as diverse as the space of all possible AI products.
- Yes, certain generic RL envs will rapidly commoditize ('web browsing', 'computer OS') - but these are not the useful ones anyway!
- The useful RL envs will be deeply domain- and product-specific – and will require corresponding human expertise and customization to build and evolve over time.

>> RL (and RL envs) will be all that you need!
- Current evidence suggests that RL / RL envs will be one part of the overall AI development loop- which will continue to require golden human annotations/traces for initial SFT; ongoing human evals; and more
- Just like trial-and-error based learning is only one part of human learning, RL will likely be one tool/phase of many.

In summary:
- (1) Building the components of an RL environment is usually highly non-trivial.
- (2) RL envs effectively describe a product spec - there will be a wide range of unique ones, requiring deep product/domain expertise.
- (3) RL (and RL envs) will be one component of a rich ecosystem of tools for model learning, including human data, rubrics, evals, and more.

If interested in some of the work the @SnorkelAI team is doing in partnership with leading LLM developers here- shoot us a note!

It's an exciting time to build in this space :)

[Friea Berg @friea]

@nickaturley Just sent to my college freshman - these are really, really good! Thank you for sharing

[Friea Berg @friea]

@embirico @tbpn listening to the Apple discussion and looking forward to hearing the latest on GPT‑5-Codex

[Friea Berg @friea]

@united help! I’m stuck in Malaga trying to get home to California. Last night’s flight was cancelled, today’s rebooked flight is delayed on outbound to Newark. System rebooked me & split tickets. Local gate can’t help

[Friea Berg @friea]

help! I’m stuck in Malaga trying to get home to California. Last night’s flight was cancelled, today’s rebooked @united flight is delayed on outbound to Newark. System rebooked me & split tickets. Local gate can’t help.

[Alex Ratner @ajratner]

1/ Super excited to deepen our partnership with @Azure! Most real-world use cases end up being blocked on the data. Solve data-centric development with @SnorkelAI , then connect seamlessly to Azure AI for model development and serving: https://t.co/1dZhPISexb

[Holger Müller #EnterpriseAcceleration @holgermu]

MyPOV - You know the #AI / #ML transformation has hit the mainstream when... a European / German (fair - high tech) company - presents in the US... #SplunkConf

[𝙲𝚘𝚛𝚢 𝙼𝚒𝚗𝚝𝚘𝚗 @Cory_Minton]

"Travel is the cure to ignorance."

Brilliant quote from @CarnivalCruise CEO at @Splunkconf #keynote.

[Friea Berg @friea]

#deserate4coffee

[Matthias Maier @Matthias_BY]

Starbucks using Splunk ES for years and together with Phantom over 2 Years! Please continue to protect my coffee ☕️ yummy waiting for a break to get one #splunkconf18 #siem #protectthebusiness

[Friea Berg @friea]

@sfmta_muni @NorthBeachNow @SRESproductions @SFGate @sfchronicle @ESernoffsky Watching the traffic fiasco at Kearny & Vallejo due to poor @sfmta_muni office placement. Sigh.

[Friea Berg @friea]

SF North Beach festival - @sfmta_muni officer ignores oncoming traffic and illegal u-turns to enjoy the sun. @NorthBeachNow @SRESproductions pls fix this next year #abc7now @sfgate @sfchronicle @esernoffsky

[Friea Berg @friea]

SF North Beach festival - officer ignores oncoming traffic and illegal u-turns to enjoy the sun. @NorthBeachNow pls fix this next year