Buy Crypto
Pay with
USD
USD
Buy & Sell
Hot
Supports Visa, Mastercard, SEPA & more
P2P
0 Fees
Flexible trading, zero fees
Gate Card
Use your crypto for payments worldwide
Markets
Trade
Basic
Advanced
Futures
Futures
Access hundreds of perpetual contracts
CFD
Gold
One platform for global traditional assets
Options
Hot
Trade European-style vanilla options
Unified Account
Maximize your capital efficiency
Demo Trading
Introduction to Futures Trading
Learn the basics of futures trading
Futures Events
Join events to earn rewards
Demo Trading
Use virtual funds to practice risk-free trading
Earn
Launch
CandyDrop
Collect candies to earn airdrops
Launchpool
Quick staking, earn potential new tokens
HODLer Airdrop
Hold GT and get massive airdrops for free
Pre-IPOs
Unlock full access to global stock IPOs
Alpha Points
Trade on-chain assets and earn airdrops
Futures Points
Earn futures points and claim airdrop rewards
Investment
Square
Square
Post, share, and explore crypto trends
Live
moments live
Live crypto market analysis
Chat
Chat with crypto traders
News
What is happening in crypto
flower_head
More
Promotions
AI
Others
CFD
Rewards
Square
Latest
Hot
News
Profile

Stanford NLP: Most publicly available agent training data still concentrates on the post-training phase

MeNews
robot
Abstract generation in progress
AIMPACT message, May 15 (UTC+8), Stanford NLP team pointed out on Twitter that most publicly available agentic training data still mainly focuses on the post-training phase, especially for models like Qwen (which may have already been trained on large amounts of agent data). The team believes that to train a good open-source model from scratch, the amount of agent data required far exceeds what is needed for just post-training from open weights, highlighting the current insufficiency of agent training data during the pre-training phase. (Source: InFoQ)
View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • 12
  • 8
  • Share
Comment
Add a comment
Add a comment
SushiAndSlugs
· 4h ago
Does this count as an easy win for Qwen?
View OriginalReply0
FlamingoFacingJudgment
· 5h ago
Open-source models want to pursue closed-source, data barriers are harder to break than computing power
View OriginalReply0
ColdLightNftCabinet
· 8h ago
Open source communities need to think about how to crowdfund pre-training data.
View OriginalReply0
GateUser-a365d15f
· 8h ago
It feels like we're back to the old story that data is power.
View OriginalReply0
GateUser-46033407
· 8h ago
The amount of data needed to train from zero sounds hopeless.
View OriginalReply0
PerpNightshift
· 9h ago
This research has handed a weapon to the closed-source camp.
View OriginalReply0
GateUser-46c777d0
· 9h ago
Stanford's observation is very accurate; agent capabilities are indeed built up through training in later stages.
View OriginalReply0
GlassDomeRoaming
· 9h ago
There is always a limit to post-training optimization, and the shortcomings of pre-training will eventually be exposed.
View OriginalReply0
GateUser-e84f640c
· 9h ago
This conclusion is quite discouraging for small and medium teams; the data threshold is getting higher and higher.
View OriginalReply0
ExitLiquidityStan
· 9h ago
I hope someone can open-source some high-quality pre-trained agent data.
View OriginalReply0
View More

  • Pinned

Sitemap