#deepswe - Twitter Hashtag

2 months ago

Yapay zeka veri inovasyon şirketi Datacurve, otonom kodlama modellerini mevcut halka açık liderlik tablolarından çok daha doğru değerlendirmek üzere tasarlanan yeni DeepSWE kriter testini paylaştı. #Datacurve #DeepSWE #CodingAI #Benchmark #TechNews https://t.co/mNZXyCPUC7

0

4

0

1

85

ChillTA🪙

@chilltc

2 months ago

Question for legal AI partners using #Anthropic: after #DeepSWE, who audited the benchmarks you were shown? If coding evals rewarded answer-key mining from git history, what checked legal evals for leakage, stale law, fake citations and retrieval failure? Who knew what, and when? #LegalAI #LegalTech #AIGovernance #DeepSWE #Benchmarking #Anthropic

chilltc's tweet photo. Question for legal AI partners using #Anthropic: after #DeepSWE, who audited the benchmarks you were shown? If coding evals rewarded answer-key mining from git history, what checked legal evals for leakage, stale law, fake citations and retrieval failure? Who knew what, and when?
#LegalAI #LegalTech #AIGovernance #DeepSWE #Benchmarking #Anthropic

0

2

0

70

ChillTA🪙

@chilltc

2 months ago

Bad Slot for Anthropic... New benchmark from Datacurve (#DeepSWE ) just rearranged the AI coding leaderboard. • GPT-5.5: 70% • GPT-5.4: 56% • Claude Opus 4.7: 54% • #Claude Haiku 4.5: 0% (was 39% on SWE-Bench Pro) The real story isn't the ranking. It's that ~18% of Claude #Opus 4.7's SWE-Bench Pro "passes" were the agent running `git log --all` and copying the merged fix out of git history. If your model selection is downstream of a leaderboard, the leaderboard needs to be downstream of an audit. #AI #Benchmark #Anthropic #OpenAi

chilltc's tweet photo. Bad Slot for Anthropic... New benchmark from Datacurve (#DeepSWE ) just rearranged the AI coding leaderboard.

• GPT-5.5: 70%
• GPT-5.4: 56%
• Claude Opus 4.7: 54%
• #Claude Haiku 4.5: 0% (was 39% on SWE-Bench Pro)

The real story isn't the ranking. It's that ~18% of Claude #Opus 4.7's SWE-Bench Pro "passes" were the agent running `git log --all` and copying the merged fix out of git history.

If your model selection is downstream of a leaderboard, the leaderboard needs to be downstream of an audit.
#AI #Benchmark #Anthropic #OpenAi

0

2

0

1K

Arman

@armanzeroeight

2 months ago

every #AI coding benchmark has the same problem: the models have already seen the answers. #SWE-Bench scrapes real GitHub PRs. those PRs are public. they were in training data. #DeepSWE does something different. here's what:

armanzeroeight's tweet photo. every #AI coding benchmark has the same problem:
the models have already seen the answers.

#SWE-Bench scrapes real GitHub PRs. those PRs are public. they were in training data.

#DeepSWE does something different. here's what: https://t.co/HgpD8Y7Mvw

2

1

0

161

Jaskirat Singh

@1jaskiratsingh

10 months ago

Excited to be #COLM2025 🇨🇦. Would love to meet old and new friends!!! Recently I have been thinking how we can bootstrap rich VQA data for better scaling RL with non-verifiable rewards. Hit me up if you want to chat about scaling reasoning driven SWE agents #R2EGym #DeepSWE We will be presenting #R2EGym on Wed morning 👇

Naman Jain

@StringChaos

over 1 year ago

Excited to release R2E-Gym - 🔥 8.1K executable environments using synthetic data - 🧠 Hybrid verifiers for enhanced inference-time scaling - 📈 51% success-rate on the SWE-Bench Verified - 🤗 Open Source Data + Models + Trajectories 1/

StringChaos's tweet photo. Excited to release R2E-Gym
- 🔥 8.1K executable environments using synthetic data
- 🧠 Hybrid verifiers for enhanced inference-time scaling
- 📈 51% success-rate on the SWE-Bench Verified
- 🤗 Open Source Data + Models + Trajectories

1/ https://t.co/qjXiApIffy

16

256

60

115

52K

0

15

3

4

5K

AI Gateway @AIGatewayEN

about 1 year ago

🏭Industry&Tech 🤖#Baidu's Ernie teams up with #Xiaomi App Store—AI agents go mainstream. 💻Shortcut debuts! #AI Excel assistant outperforms human champs 10x. 🔍#Bilibili open-sources AniSora V3—generate anime videos in one click. 🦾#DeepSWE AI Agent system tops charts—powered by Qwen3-32B. 🚗#XPeng G7 Ultra drops—with next-gen autonomous driving model.

AIGatewayEN's tweet photo. 🏭Industry&Tech

🤖#Baidu's Ernie teams up with #Xiaomi App Store—AI agents go mainstream.
💻Shortcut debuts! #AI Excel assistant outperforms human champs 10x.
🔍#Bilibili open-sources AniSora V3—generate anime videos in one click.
🦾#DeepSWE AI Agent system tops charts—powered by Qwen3-32B.
🚗#XPeng G7 Ultra drops—with next-gen autonomous driving model.

1

2

0

164

Top Tweets for #deepswe

Last Seen Hashtags on Sotwe

Trends for you

Most Popular Users