🤖 Agentic AI: Agents now plan and use tools.
📊 Mega-Models: Muse Spark (816B), MAI-Base-2 (2.8T).
⚡ Fast Inference: DeepInfer boosts speed, cuts costs.
#AI2026 #AgenticAI #MegaModels #FastAI
View in Timelines
📊 Mega-Models: Muse Spark (816B), MAI-Base-2 (2.8T).
⚡ Fast Inference: DeepInfer boosts speed, cuts costs.
#AI2026 #AgenticAI #MegaModels #FastAI
View in Timelines
October 21, 2026 at 2:01 PM
🤖 Agentic AI: Agents now plan and use tools.
📊 Mega-Models: Muse Spark (816B), MAI-Base-2 (2.8T).
⚡ Fast Inference: DeepInfer boosts speed, cuts costs.
#AI2026 #AgenticAI #MegaModels #FastAI
View in Timelines
📊 Mega-Models: Muse Spark (816B), MAI-Base-2 (2.8T).
⚡ Fast Inference: DeepInfer boosts speed, cuts costs.
#AI2026 #AgenticAI #MegaModels #FastAI
View in Timelines
🤖 Open models: Beam & DeepSeek V4.1-Flash released.
🕹️ Smarter agents: Harness² enables AI self-optimization.
⚡ Fast gen: Mercury 3 exceeds 1,000 tokens/sec.
#AITrends2026 #OpenAI #SelfImprovingAI #FastAI
#AITrends2026 #OpenAI #SelfImprovingAI #FastAI
View in Timelines
🕹️ Smarter agents: Harness² enables AI self-optimization.
⚡ Fast gen: Mercury 3 exceeds 1,000 tokens/sec.
#AITrends2026 #OpenAI #SelfImprovingAI #FastAI
#AITrends2026 #OpenAI #SelfImprovingAI #FastAI
View in Timelines
October 20, 2026 at 2:01 PM
🤖 Open models: Beam & DeepSeek V4.1-Flash released.
🕹️ Smarter agents: Harness² enables AI self-optimization.
⚡ Fast gen: Mercury 3 exceeds 1,000 tokens/sec.
#AITrends2026 #OpenAI #SelfImprovingAI #FastAI
#AITrends2026 #OpenAI #SelfImprovingAI #FastAI
View in Timelines
🕹️ Smarter agents: Harness² enables AI self-optimization.
⚡ Fast gen: Mercury 3 exceeds 1,000 tokens/sec.
#AITrends2026 #OpenAI #SelfImprovingAI #FastAI
#AITrends2026 #OpenAI #SelfImprovingAI #FastAI
View in Timelines
今日fastaiの6: Random forests終わらせよ
September 15, 2026 at 3:38 AM
今日fastaiの6: Random forests終わらせよ
fastai急に飽きたから捨ててゼロ作本やってたけど、楽しくてご機嫌🎶やっぱり作るって楽しい🎶
September 15, 2026 at 6:33 AM
fastai急に飽きたから捨ててゼロ作本やってたけど、楽しくてご機嫌🎶やっぱり作るって楽しい🎶
fastaiあと2レッスンだからザッと終わらせて、ゼロ作とKarpathyやったら解釈可能性に突っ込め
September 14, 2026 at 2:58 PM
fastaiあと2レッスンだからザッと終わらせて、ゼロ作とKarpathyやったら解釈可能性に突っ込め
fastaiめちゃくちゃ勉強になるけれど、疲れる
September 14, 2026 at 8:58 AM
fastaiめちゃくちゃ勉強になるけれど、疲れる
今日やる
・掃除
・fastaiコード読み
・aw読み＆感想
・ドラマ見る
・読書
・絵描く
・掃除
・fastaiコード読み
・aw読み＆感想
・ドラマ見る
・読書
・絵描く
September 12, 2026 at 1:34 AM
今日やる
・掃除
・fastaiコード読み
・aw読み＆感想
・ドラマ見る
・読書
・絵描く
・掃除
・fastaiコード読み
・aw読み＆感想
・ドラマ見る
・読書
・絵描く
fastaiってやっぱ2週目やった方が良いんだ
August 30, 2026 at 5:11 PM
fastaiってやっぱ2週目やった方が良いんだ
The pack uses 5,133 completed FBS games and 86 documented fields.
The workflow moves from ridge and random forest baselines through logistic regression, XGBoost, FastAI, SHAP interpretation, and a stacked ensemble.
The workflow moves from ridge and random forest baselines through logistic regression, XGBoost, FastAI, SHAP interpretation, and a stacked ensemble.
August 6, 2026 at 4:03 PM
The pack uses 5,133 completed FBS games and 86 documented fields.
The workflow moves from ridge and random forest baselines through logistic regression, XGBoost, FastAI, SHAP interpretation, and a stacked ensemble.
The workflow moves from ridge and random forest baselines through logistic regression, XGBoost, FastAI, SHAP interpretation, and a stacked ensemble.
From pack to predictions in minutes.
New guide shows you:
• Which CFBD API endpoints contain the features you need
• How to load & run sklearn, XGBoost, and fastai models
• How Tier 3 weekly CSV drops give you ready-to-score data instantly (week 5+)
📖 buff.ly/jaEKQcb
New guide shows you:
• Which CFBD API endpoints contain the features you need
• How to load & run sklearn, XGBoost, and fastai models
• How Tier 3 weekly CSV drops give you ready-to-score data instantly (week 5+)
📖 buff.ly/jaEKQcb
From Model Training Pack to Predictions in Minutes
This easy guide shows you how to load your Model Training Pack models, get the right data, and start predicting, plus how Tier 3 weekly CSV drops make it instant.
blog.collegefootballdata.com
August 12, 2025 at 1:13 AM
From pack to predictions in minutes.
New guide shows you:
• Which CFBD API endpoints contain the features you need
• How to load & run sklearn, XGBoost, and fastai models
• How Tier 3 weekly CSV drops give you ready-to-score data instantly (week 5+)
📖 buff.ly/jaEKQcb
New guide shows you:
• Which CFBD API endpoints contain the features you need
• How to load & run sklearn, XGBoost, and fastai models
• How Tier 3 weekly CSV drops give you ready-to-score data instantly (week 5+)
📖 buff.ly/jaEKQcb
✅ Opponent-adjusted features
✅ Week 5+ only. no early-season noise.
✅ Targets: margin, win prob, team points
✅ 7 ML notebooks (XGBoost, fastai, SHAP + more)
🗓️ Weekly updates start Week 5 for Patreon Tier 3+
👉 buff.ly/5kjPhWt
✅ Week 5+ only. no early-season noise.
✅ Targets: margin, win prob, team points
✅ 7 ML notebooks (XGBoost, fastai, SHAP + more)
🗓️ Weekly updates start Week 5 for Patreon Tier 3+
👉 buff.ly/5kjPhWt
August 1, 2025 at 2:01 PM
✅ Opponent-adjusted features
✅ Week 5+ only. no early-season noise.
✅ Targets: margin, win prob, team points
✅ 7 ML notebooks (XGBoost, fastai, SHAP + more)
🗓️ Weekly updates start Week 5 for Patreon Tier 3+
👉 buff.ly/5kjPhWt
✅ Week 5+ only. no early-season noise.
✅ Targets: margin, win prob, team points
✅ 7 ML notebooks (XGBoost, fastai, SHAP + more)
🗓️ Weekly updates start Week 5 for Patreon Tier 3+
👉 buff.ly/5kjPhWt
🚨 Drops tomorrow: The Model Training Pack
✅ Prebuilt CFB training data (week 5+; opponent-adjusted)
✅ ML-ready targets: win prob, score margin, team points
✅ 7 notebooks: XGBoost, fastai, SHAP, stacking + more
Starter Pack buyers: discount codes hit inboxes tomorrow 👀
#CFB #SportsAnalytics
✅ Prebuilt CFB training data (week 5+; opponent-adjusted)
✅ ML-ready targets: win prob, score margin, team points
✅ 7 notebooks: XGBoost, fastai, SHAP, stacking + more
Starter Pack buyers: discount codes hit inboxes tomorrow 👀
#CFB #SportsAnalytics
July 31, 2025 at 3:30 PM
🚨 Drops tomorrow: The Model Training Pack
✅ Prebuilt CFB training data (week 5+; opponent-adjusted)
✅ ML-ready targets: win prob, score margin, team points
✅ 7 notebooks: XGBoost, fastai, SHAP, stacking + more
Starter Pack buyers: discount codes hit inboxes tomorrow 👀
#CFB #SportsAnalytics
✅ Prebuilt CFB training data (week 5+; opponent-adjusted)
✅ ML-ready targets: win prob, score margin, team points
✅ 7 notebooks: XGBoost, fastai, SHAP, stacking + more
Starter Pack buyers: discount codes hit inboxes tomorrow 👀
#CFB #SportsAnalytics
📊 A closer look at the Model Training Pack (drops Friday)
✅ Opponent-adjusted features
✅ Prebuilt models: XGBoost, fastai, SHAP, more
✅ Realistic targets: win prob, score margin
Built to help you go from idea → model → insight
#CFB #SportsAnalytics #MachineLearning
✅ Opponent-adjusted features
✅ Prebuilt models: XGBoost, fastai, SHAP, more
✅ Realistic targets: win prob, score margin
Built to help you go from idea → model → insight
#CFB #SportsAnalytics #MachineLearning
July 30, 2025 at 3:45 PM
📊 A closer look at the Model Training Pack (drops Friday)
✅ Opponent-adjusted features
✅ Prebuilt models: XGBoost, fastai, SHAP, more
✅ Realistic targets: win prob, score margin
Built to help you go from idea → model → insight
#CFB #SportsAnalytics #MachineLearning
✅ Opponent-adjusted features
✅ Prebuilt models: XGBoost, fastai, SHAP, more
✅ Realistic targets: win prob, score margin
Built to help you go from idea → model → insight
#CFB #SportsAnalytics #MachineLearning
In one of the earliest fastai courses Jeremy Howard casually made an artificial neural network with Excel. Fun stuff
October 3, 2025 at 2:31 PM
In one of the earliest fastai courses Jeremy Howard casually made an artificial neural network with Excel. Fun stuff
I posted a couple times about the fastai course providing insights into how AI models can get worse on synthetic data a long while ago, so I find it interesting that one of the main authors (Jeremy Howard) seems to have reservations about the model collapse paper
September 5, 2024 at 9:05 PM
I posted a couple times about the fastai course providing insights into how AI models can get worse on synthetic data a long while ago, so I find it interesting that one of the main authors (Jeremy Howard) seems to have reservations about the model collapse paper
On training through AI generated input: I gotta wonder how many devs in the industry had the same insights as what you would find in like... the first few chapters of the free fastai course 7-8 years ago (among others), and gasped when the floodgates got opened. This was absolutely predictable
This has very important implications on the *infinite* amounts of energy that AI slop generators may need, were they to figure out a way to make this work and train them through AI generated input.
August 25, 2024 at 10:02 PM
On training through AI generated input: I gotta wonder how many devs in the industry had the same insights as what you would find in like... the first few chapters of the free fastai course 7-8 years ago (among others), and gasped when the floodgates got opened. This was absolutely predictable
I'm glad I learned about the guts of AI models through the *free* fastAI course because they have been hammering this point first thing since *at least* 2017
Shocker - ChatGPT and other LLMs/similar AI models collapse in on themselves like a flan in a cupboard when they start eating their own data.
AI models collapse when trained on recursively generated data - Nature
Analysis shows that indiscriminately training generative artificial intelligence on real and generated content, usually done by scraping data from the Internet, can lead to a collapse in th...
www.nature.com
July 24, 2024 at 10:49 PM
I'm glad I learned about the guts of AI models through the *free* fastAI course because they have been hammering this point first thing since *at least* 2017
Worryingly this seems very doable. This is the kind of thing folks like Rachel Thomas (fastai) were warning about like 7 or 8 years ago
New: Trump-appointed Judge John Bush of the 6th Circuit today in a speech to a Federalist Society chapter laid out a vision for how future AI technology could make it even easier for originalist judges to support rulings based on "history and tradition."
Conservative US judge says AI could strengthen 'originalist' movement
A conservative federal appeals court judge on Monday said future developments in artificial intelligence could revolutionize how judges like himself tie their rulings to the past by interpreting the U.S. Constitution based on the nation's history and tradition.
www.reuters.com
April 1, 2024 at 7:53 PM
Worryingly this seems very doable. This is the kind of thing folks like Rachel Thomas (fastai) were warning about like 7 or 8 years ago
The newly lauched solveit (solveit.fast.ai) from @howard.fm /FastAI addresses this, with the focus on us learning better
Solve It With Code
solveit.fast.ai
August 22, 2025 at 4:05 AM
The newly lauched solveit (solveit.fast.ai) from @howard.fm /FastAI addresses this, with the focus on us learning better
Resulta que si pintas de una cierta manera los "embeddings" (vectores que representan palabras) que corresponden a nombres de estados alemanes, puedes obtener una aproximación del mapa de Alemania. Fuente: github.com/fastai/fastb...
November 17, 2024 at 4:24 PM
Resulta que si pintas de una cierta manera los "embeddings" (vectores que representan palabras) que corresponden a nombres de estados alemanes, puedes obtener una aproximación del mapa de Alemania. Fuente: github.com/fastai/fastb...
started working through the fastai ML course today! i'm pumped
June 3, 2026 at 1:52 AM
started working through the fastai ML course today! i'm pumped
Enjoyed this podcast from Jeremy Howard, on finding your own path in life, ULMFit, what's wrong with language models, Mojo, etc. Interesting character www.latent.space/p/fastai
November 20, 2023 at 9:57 PM
Enjoyed this podcast from Jeremy Howard, on finding your own path in life, ULMFit, what's wrong with language models, Mojo, etc. Interesting character www.latent.space/p/fastai
I could imagine the need to reverse engineer a more fastai approach to teaching the same content in the context of a modern analysis workflow.
October 3, 2023 at 8:57 PM
I could imagine the need to reverse engineer a more fastai approach to teaching the same content in the context of a modern analysis workflow.