MadAIMan
@mayankvora.bsky.social
AI Newsletter Pioneer at The Shift | Get the latest breakthroughs, tools, and strategies in AI | Subscribe to my newsletter 👇🏻
https://theshiftai.beehiiv.com/subscribe
https://theshiftai.beehiiv.com/subscribe
AI is not going to take job.
Our newsletter, The Shift, delivers breakthroughs, tools, and strategies to help you become value creator and build in this new era easily.
Subscribe: theshiftai.beehiiv.com/subscribe
Plus: Get access to 2k+ AI Tools and free AI courses when you join.
Our newsletter, The Shift, delivers breakthroughs, tools, and strategies to help you become value creator and build in this new era easily.
Subscribe: theshiftai.beehiiv.com/subscribe
Plus: Get access to 2k+ AI Tools and free AI courses when you join.
The Shift
Get the latest breakthroughs, tools, and strategies to level up at your work all in under 5 minutes a day, delivered 5 days a week.
theshiftai.beehiiv.com
December 8, 2025 at 2:05 PM
AI is not going to take job.
Our newsletter, The Shift, delivers breakthroughs, tools, and strategies to help you become value creator and build in this new era easily.
Subscribe: theshiftai.beehiiv.com/subscribe
Plus: Get access to 2k+ AI Tools and free AI courses when you join.
Our newsletter, The Shift, delivers breakthroughs, tools, and strategies to help you become value creator and build in this new era easily.
Subscribe: theshiftai.beehiiv.com/subscribe
Plus: Get access to 2k+ AI Tools and free AI courses when you join.
3/ Scientists want AI to propose hypotheses and design experiments, but trust is low, so usage stays stuck in writing and debugging.
The real AI gap is not adoption, it’s legitimacy, until teams normalize it, the biggest ROI stays trapped in whispers.
The real AI gap is not adoption, it’s legitimacy, until teams normalize it, the biggest ROI stays trapped in whispers.
December 8, 2025 at 2:05 PM
3/ Scientists want AI to propose hypotheses and design experiments, but trust is low, so usage stays stuck in writing and debugging.
The real AI gap is not adoption, it’s legitimacy, until teams normalize it, the biggest ROI stays trapped in whispers.
The real AI gap is not adoption, it’s legitimacy, until teams normalize it, the biggest ROI stays trapped in whispers.
2/ Creatives are split. They use AI for ideation and output velocity, then hide it to protect identity and credibility. The tension is not capability, it’s ownership, originality, and social permission.
December 8, 2025 at 2:05 PM
2/ Creatives are split. They use AI for ideation and output velocity, then hide it to protect identity and credibility. The tension is not capability, it’s ownership, originality, and social permission.
1/ Workers love the speed. 86% say AI saves them time. But 69% still feel stigma using it at work, and over half carry job anxiety about what happens when “help” becomes “replacement.”
December 8, 2025 at 2:05 PM
1/ Workers love the speed. 86% say AI saves them time. But 69% still feel stigma using it at work, and over half carry job anxiety about what happens when “help” becomes “replacement.”
AI is not going to take job.
Our newsletter, The Shift, delivers breakthroughs, tools, and strategies to help you become value creator and build in this new era easily.
Subscribe: theshiftai.beehiiv.com/subscribe
Plus: Get access to 2k+ AI Tools and free AI courses when you join.
Our newsletter, The Shift, delivers breakthroughs, tools, and strategies to help you become value creator and build in this new era easily.
Subscribe: theshiftai.beehiiv.com/subscribe
Plus: Get access to 2k+ AI Tools and free AI courses when you join.
The Shift
Get the latest breakthroughs, tools, and strategies to level up at your work all in under 5 minutes a day, delivered 5 days a week.
theshiftai.beehiiv.com
December 4, 2025 at 2:10 PM
AI is not going to take job.
Our newsletter, The Shift, delivers breakthroughs, tools, and strategies to help you become value creator and build in this new era easily.
Subscribe: theshiftai.beehiiv.com/subscribe
Plus: Get access to 2k+ AI Tools and free AI courses when you join.
Our newsletter, The Shift, delivers breakthroughs, tools, and strategies to help you become value creator and build in this new era easily.
Subscribe: theshiftai.beehiiv.com/subscribe
Plus: Get access to 2k+ AI Tools and free AI courses when you join.
3/ This is “deployment beats hype.” When API outages and compliance block closed models, open weights win by default. Expect more laptops, edge devices, vehicles, robots, and on prem stacks to standardize on small, tuned models over giant black boxes.
December 4, 2025 at 2:10 PM
3/ This is “deployment beats hype.” When API outages and compliance block closed models, open weights win by default. Expect more laptops, edge devices, vehicles, robots, and on prem stacks to standardize on small, tuned models over giant black boxes.
2/ The real story is Ministral. 3B, 8B, 14B in base, instruct, and reasoning, Apache 2.0, vision-capable, 128k to 256k context, runs on a single GPU or fully offline.
December 4, 2025 at 2:10 PM
2/ The real story is Ministral. 3B, 8B, 14B in base, instruct, and reasoning, Apache 2.0, vision-capable, 128k to 256k context, runs on a single GPU or fully offline.
1/ Large 3 is the flex. 41B active MoE, multimodal plus multilingual, 256k context, built for coding, docs, assistants, and automation without feeling like a toy.
December 4, 2025 at 2:10 PM
1/ Large 3 is the flex. 41B active MoE, multimodal plus multilingual, 256k context, built for coding, docs, assistants, and automation without feeling like a toy.
The real kicker, open weights.
- Both 685B models ship MIT licensed with weights on Hugging Face, so teams can self host, fine tune, and own the stack.
- Both 685B models ship MIT licensed with weights on Hugging Face, so teams can self host, fine tune, and own the stack.
December 2, 2025 at 1:59 PM
The real kicker, open weights.
- Both 685B models ship MIT licensed with weights on Hugging Face, so teams can self host, fine tune, and own the stack.
- Both 685B models ship MIT licensed with weights on Hugging Face, so teams can self host, fine tune, and own the stack.
3/ Breakthrough 3, pricing that breaks budgets.
- V3.2 is priced at $0.28 input and $0.42 output per million tokens, which undercuts most flagship reasoning APIs.
- That means startups can run heavier agent workflows without getting rate limited by cost.
- V3.2 is priced at $0.28 input and $0.42 output per million tokens, which undercuts most flagship reasoning APIs.
- That means startups can run heavier agent workflows without getting rate limited by cost.
December 2, 2025 at 1:59 PM
3/ Breakthrough 3, pricing that breaks budgets.
- V3.2 is priced at $0.28 input and $0.42 output per million tokens, which undercuts most flagship reasoning APIs.
- That means startups can run heavier agent workflows without getting rate limited by cost.
- V3.2 is priced at $0.28 input and $0.42 output per million tokens, which undercuts most flagship reasoning APIs.
- That means startups can run heavier agent workflows without getting rate limited by cost.
2/ Breakthrough 2, RL scaling.
- More post training compute pushes V3.2 toward frontier performance, and Speciale claims to reach or beat top tier reasoning.
- They also claim validation via 2025 IMO and IOI gold medal level results.
- More post training compute pushes V3.2 toward frontier performance, and Speciale claims to reach or beat top tier reasoning.
- They also claim validation via 2025 IMO and IOI gold medal level results.
December 2, 2025 at 1:59 PM
2/ Breakthrough 2, RL scaling.
- More post training compute pushes V3.2 toward frontier performance, and Speciale claims to reach or beat top tier reasoning.
- They also claim validation via 2025 IMO and IOI gold medal level results.
- More post training compute pushes V3.2 toward frontier performance, and Speciale claims to reach or beat top tier reasoning.
- They also claim validation via 2025 IMO and IOI gold medal level results.
1/ Breakthrough 1, Sparse Attention.
- DeepSeek’s DSA cuts long context compute while keeping accuracy, so math, code, and deep reasoning stop being a hardware tax.
- Translation, this scales without lighting money on fire.
- DeepSeek’s DSA cuts long context compute while keeping accuracy, so math, code, and deep reasoning stop being a hardware tax.
- Translation, this scales without lighting money on fire.
December 2, 2025 at 1:59 PM
1/ Breakthrough 1, Sparse Attention.
- DeepSeek’s DSA cuts long context compute while keeping accuracy, so math, code, and deep reasoning stop being a hardware tax.
- Translation, this scales without lighting money on fire.
- DeepSeek’s DSA cuts long context compute while keeping accuracy, so math, code, and deep reasoning stop being a hardware tax.
- Translation, this scales without lighting money on fire.
AI is not going to take job.
Our newsletter, The Shift, delivers breakthroughs, tools, and strategies to help you become value creator and build in this new era easily.
Subscribe: theshiftai.beehiiv.com/subscribe
Plus: Get access to 2k+ AI Tools and free AI courses when you join.
Our newsletter, The Shift, delivers breakthroughs, tools, and strategies to help you become value creator and build in this new era easily.
Subscribe: theshiftai.beehiiv.com/subscribe
Plus: Get access to 2k+ AI Tools and free AI courses when you join.
The Shift
Get the latest breakthroughs, tools, and strategies to level up at your work all in under 5 minutes a day, delivered 5 days a week.
theshiftai.beehiiv.com
December 1, 2025 at 2:37 PM
AI is not going to take job.
Our newsletter, The Shift, delivers breakthroughs, tools, and strategies to help you become value creator and build in this new era easily.
Subscribe: theshiftai.beehiiv.com/subscribe
Plus: Get access to 2k+ AI Tools and free AI courses when you join.
Our newsletter, The Shift, delivers breakthroughs, tools, and strategies to help you become value creator and build in this new era easily.
Subscribe: theshiftai.beehiiv.com/subscribe
Plus: Get access to 2k+ AI Tools and free AI courses when you join.
The U.S. leads in frontier research.
China now leads in global adoption, influence, and the open ecosystem that shapes thousands of apps.
China now leads in global adoption, influence, and the open ecosystem that shapes thousands of apps.
December 1, 2025 at 2:37 PM
The U.S. leads in frontier research.
China now leads in global adoption, influence, and the open ecosystem that shapes thousands of apps.
China now leads in global adoption, influence, and the open ecosystem that shapes thousands of apps.
4/ Transparency collapses
- Open-source disclosure fell from 79% → 39% in 3 years.
- Chinese models increasingly embed state-aligned narratives.
- Open-source disclosure fell from 79% → 39% in 3 years.
- Chinese models increasingly embed state-aligned narratives.
December 1, 2025 at 2:37 PM
4/ Transparency collapses
- Open-source disclosure fell from 79% → 39% in 3 years.
- Chinese models increasingly embed state-aligned narratives.
- Open-source disclosure fell from 79% → 39% in 3 years.
- Chinese models increasingly embed state-aligned narratives.
3/ Efficiency becomes China’s edge
- Chip bans forced innovation: distillation, compression, hardware-light architectures.
- Perfect for Asia, LATAM, and Africa where consumer devices rule.
- Chip bans forced innovation: distillation, compression, hardware-light architectures.
- Perfect for Asia, LATAM, and Africa where consumer devices rule.
December 1, 2025 at 2:37 PM
3/ Efficiency becomes China’s edge
- Chip bans forced innovation: distillation, compression, hardware-light architectures.
- Perfect for Asia, LATAM, and Africa where consumer devices rule.
- Chip bans forced innovation: distillation, compression, hardware-light architectures.
- Perfect for Asia, LATAM, and Africa where consumer devices rule.
2/ Two strategies, two futures
- China: ship small, efficient, open models at scale.
- U.S.: lock frontier models behind APIs + enterprise paywalls.
- China: ship small, efficient, open models at scale.
- U.S.: lock frontier models behind APIs + enterprise paywalls.
December 1, 2025 at 2:37 PM
2/ Two strategies, two futures
- China: ship small, efficient, open models at scale.
- U.S.: lock frontier models behind APIs + enterprise paywalls.
- China: ship small, efficient, open models at scale.
- U.S.: lock frontier models behind APIs + enterprise paywalls.
1/ China takes the global lead
- DeepSeek + Qwen push China to 17% of all downloads.
- The U.S. drops to 15.8%, down from 40%+ when models were open.
- DeepSeek + Qwen push China to 17% of all downloads.
- The U.S. drops to 15.8%, down from 40%+ when models were open.
December 1, 2025 at 2:37 PM
1/ China takes the global lead
- DeepSeek + Qwen push China to 17% of all downloads.
- The U.S. drops to 15.8%, down from 40%+ when models were open.
- DeepSeek + Qwen push China to 17% of all downloads.
- The U.S. drops to 15.8%, down from 40%+ when models were open.
AI is not going to take job.
Our newsletter, The Shift, delivers breakthroughs, tools, and strategies to help you become value creator and build in this new era easily.
Subscribe: theshiftai.beehiiv.com/subscribe
Plus: Get access to 2k+ AI Tools and free AI courses when you join.
Our newsletter, The Shift, delivers breakthroughs, tools, and strategies to help you become value creator and build in this new era easily.
Subscribe: theshiftai.beehiiv.com/subscribe
Plus: Get access to 2k+ AI Tools and free AI courses when you join.
The Shift
Get the latest breakthroughs, tools, and strategies to level up at your work all in under 5 minutes a day, delivered 5 days a week.
theshiftai.beehiiv.com
November 27, 2025 at 1:40 PM
AI is not going to take job.
Our newsletter, The Shift, delivers breakthroughs, tools, and strategies to help you become value creator and build in this new era easily.
Subscribe: theshiftai.beehiiv.com/subscribe
Plus: Get access to 2k+ AI Tools and free AI courses when you join.
Our newsletter, The Shift, delivers breakthroughs, tools, and strategies to help you become value creator and build in this new era easily.
Subscribe: theshiftai.beehiiv.com/subscribe
Plus: Get access to 2k+ AI Tools and free AI courses when you join.
4/ A New Research Economy
- Controlled access for universities & companies, strict cybersecurity, new fellowships, and unified IP rules.
The play: Turn America’s scattered research assets into one AI-accelerated discovery engine.
- Controlled access for universities & companies, strict cybersecurity, new fellowships, and unified IP rules.
The play: Turn America’s scattered research assets into one AI-accelerated discovery engine.
November 27, 2025 at 1:40 PM
4/ A New Research Economy
- Controlled access for universities & companies, strict cybersecurity, new fellowships, and unified IP rules.
The play: Turn America’s scattered research assets into one AI-accelerated discovery engine.
- Controlled access for universities & companies, strict cybersecurity, new fellowships, and unified IP rules.
The play: Turn America’s scattered research assets into one AI-accelerated discovery engine.
3/ High-Impact Targets
- Fusion, drug discovery, materials, semiconductors, quantum.
- AI-run experiments shorten discovery cycles from years → weeks.
- Fusion, drug discovery, materials, semiconductors, quantum.
- AI-run experiments shorten discovery cycles from years → weeks.
November 27, 2025 at 1:40 PM
3/ High-Impact Targets
- Fusion, drug discovery, materials, semiconductors, quantum.
- AI-run experiments shorten discovery cycles from years → weeks.
- Fusion, drug discovery, materials, semiconductors, quantum.
- AI-run experiments shorten discovery cycles from years → weeks.