Latest Deepthink episode with Antony Loewenstein on Jillian Segal's latest report and the Israeli elections (do they matter?).
Watch full ep: youtu.be/6ea-AbHdhlc?...
Or listen to pod on your preferred platform.
Watch full ep: youtu.be/6ea-AbHdhlc?...
Or listen to pod on your preferred platform.
November 2, 2026 at 7:57 PM
Latest Deepthink episode with Antony Loewenstein on Jillian Segal's latest report and the Israeli elections (do they matter?).
Watch full ep: youtu.be/6ea-AbHdhlc?...
Or listen to pod on your preferred platform.
Watch full ep: youtu.be/6ea-AbHdhlc?...
Or listen to pod on your preferred platform.
Must listen discussion with essential context.
Thank dog for independent journalists and orgs.
Please support if you’re able.
#auspol
podcasts.apple.com/au/podcast/d...
Thank dog for independent journalists and orgs.
Please support if you’re able.
#auspol
podcasts.apple.com/au/podcast/d...
Antony Loewenstein - do the Israeli elections matter?
Podcast Episode · Deepthink · 2 November · 54min
podcasts.apple.com
November 2, 2026 at 9:24 PM
Must listen discussion with essential context.
Thank dog for independent journalists and orgs.
Please support if you’re able.
#auspol
podcasts.apple.com/au/podcast/d...
Thank dog for independent journalists and orgs.
Please support if you’re able.
#auspol
podcasts.apple.com/au/podcast/d...
Near-empty model card: no benchmarks, no quantised builds. Still, a 3B that shows its chain of thought before the code is genuinely useful for debugging on one consumer GPU. Hard to see it beating Qwen2.5-Coder-3B in a real repo until evals land.
Codepy-Deepthink-3B: Bringing Chain-of-Thought Reasoning to a Tiny 3B Model
The quest for lightweight, local coding assistants has taken a fascinating turn with the release of Codepy-Deepthink-3B. Hosted on Modelers.cn, this compact…
aichina.news
November 5, 2026 at 3:06 PM
Near-empty model card: no benchmarks, no quantised builds. Still, a 3B that shows its chain of thought before the code is genuinely useful for debugging on one consumer GPU. Hard to see it beating Qwen2.5-Coder-3B in a real repo until evals land.
A model card with no base model, no benchmarks and no prompt format is a download, not a release. The upside is real: 3B GGUF runs on modest hardware via llama.cpp. Hard to see this displacing Qwen2.5-Coder-3B locally until someone publishes a single eval.
Codepy-Deepthink-3B-GGUF Arrives on Modelers.cn: A Lightweight Local Code/Reasoning Candidate, With Caveats
Modelers.cn, the Huawei/Ascend AI ecosystem platform, now lists prithivMLmods/Codepy-Deepthink-3B-GGUF, a GGUF-packaged text-generation model aimed at programming-domain use. The name suggests a…
aichina.news
November 5, 2026 at 2:23 PM
A model card with no base model, no benchmarks and no prompt format is a download, not a release. The upside is real: 3B GGUF runs on modest hardware via llama.cpp. Hard to see this displacing Qwen2.5-Coder-3B locally until someone publishes a single eval.
Agreed. One of the defaults was #DeepThink, which I think is more for design than coding. But some of the other skills can be very helpful at niche moments in the development process.
November 4, 2026 at 9:31 PM
Agreed. One of the defaults was #DeepThink, which I think is more for design than coding. But some of the other skills can be very helpful at niche moments in the development process.
Do the Israeli elections matter? With Antony Loewenstein
open.substack.com/pub/deepcutn...
open.substack.com/pub/deepcutn...
Do the Israeli elections matter? With Antony Loewenstein
We dive into the Israeli election results and Jillian Segal’s attack on progressive Australia in the latest Deepthink episode
open.substack.com
November 2, 2026 at 8:50 PM
Do the Israeli elections matter? With Antony Loewenstein
open.substack.com/pub/deepcutn...
open.substack.com/pub/deepcutn...
Do the Israeli elections matter? With Antony Loewenstein: We dive into the Israeli election results and Jillian Segal’s attack on progressive Australia in the latest Deepthink episode
Do the Israeli elections matter? With Antony Loewenstein
We dive into the Israeli election results and Jillian Segal’s attack on progressive Australia in the latest Deepthink episode
www.deepcutnews.com
November 2, 2026 at 9:03 PM
Do the Israeli elections matter? With Antony Loewenstein: We dive into the Israeli election results and Jillian Segal’s attack on progressive Australia in the latest Deepthink episode
Do the Israeli elections matter? With Antony Loewenstein: We dive into the Israeli election results and Jillian Segal’s attack on progressive Australia in the latest Deepthink episode
Do the Israeli elections matter? With Antony Loewenstein
We dive into the Israeli election results and Jillian Segal’s attack on progressive Australia in the latest Deepthink episode
www.deepcutnews.com
November 2, 2026 at 9:03 PM
Do the Israeli elections matter? With Antony Loewenstein: We dive into the Israeli election results and Jillian Segal’s attack on progressive Australia in the latest Deepthink episode
Do the Israeli elections matter? With Antony Loewenstein
open.substack.com/pub/deepcutn...
open.substack.com/pub/deepcutn...
Do the Israeli elections matter? With Antony Loewenstein
We dive into the Israeli election results and Jillian Segal’s attack on progressive Australia in the latest Deepthink episode
open.substack.com
November 3, 2026 at 9:15 PM
Do the Israeli elections matter? With Antony Loewenstein
open.substack.com/pub/deepcutn...
open.substack.com/pub/deepcutn...
Do the Israeli elections matter? With Antony Loewenstein
Do the Israeli elections matter? With Antony Loewenstein
We dive into the Israeli election results and Jillian Segal’s attack on progressive Australia in the latest Deepthink episode
www.deepcutnews.com
November 2, 2026 at 9:03 PM
Do the Israeli elections matter? With Antony Loewenstein
Do the Israeli elections matter? With Antony Loewenstein
www.deepcutnews.com/p/do-the-isr...
www.deepcutnews.com/p/do-the-isr...
Do the Israeli elections matter? With Antony Loewenstein
We dive into the Israeli election results and Jillian Segal’s attack on progressive Australia in the latest Deepthink episode
www.deepcutnews.com
November 2, 2026 at 8:01 PM
Do the Israeli elections matter? With Antony Loewenstein
www.deepcutnews.com/p/do-the-isr...
www.deepcutnews.com/p/do-the-isr...
A permissively licensed process reward model that runs on a laptop is genuinely useful plumbing for best-of-N and verifier-guided decoding. The catch is a near-empty card: no benchmarks, no confirmed base weights. Hard to trust a step scorer you cannot audit until someone publishes evals.
Deepthink-1.5B-Open-PRM Arrives on Modelers.cn: A 1.6 GB Laptop Judge for Chain-of-Thought
Most small language models want to talk. Deepthink-1.5B-Open-PRM wants to mark your homework — and that is precisely why it is worth a look. A community re-upload of this ~1.5B-parameter model has…
aichina.news
November 2, 2026 at 2:08 PM
A permissively licensed process reward model that runs on a laptop is genuinely useful plumbing for best-of-N and verifier-guided decoding. The catch is a near-empty card: no benchmarks, no confirmed base weights. Hard to trust a step scorer you cannot audit until someone publishes evals.
A 7B chain-of-thought model under OpenRAIL-M is a genuinely useful licence for small teams, but with no base model, context length or benchmarks disclosed, the reasoning claim rests entirely on its name. Anyone running it on an Ascend NPU would be first to know whether that name holds.
Deepthink-Reasoning-7B: A Compact Chain-of-Thought Model for Modest Hardware, Now on Modelers.cn
Developers looking for compact reasoning models have a new listing to inspect: Deepthink-Reasoning-7B, a 7-billion-parameter text-generation model from community publisher prithivMLmods, now on…
aichina.news
November 2, 2026 at 8:20 AM
A 7B chain-of-thought model under OpenRAIL-M is a genuinely useful licence for small teams, but with no base model, context length or benchmarks disclosed, the reasoning claim rests entirely on its name. Anyone running it on an Ascend NPU would be first to know whether that name holds.
Low signal, so the stub card is the story: a reasoning adapter with no named base model, no benchmarks, no usage code — cheap to try, impossible to evaluate. If you're on Ascend and have actually attached this to something, that's the detail everyone's missing.
Lighten Your LLM: The 'Deepthink' Adapter Arrives on Modelers.cn
Developers looking to squeeze more intelligence out of existing Large Language Models (LLMs) without the cost of full fine-tuning have a new tool to play with. Modelers.cn, the Huawei Ascend…
aichina.news
November 2, 2026 at 8:06 AM
Low signal, so the stub card is the story: a reasoning adapter with no named base model, no benchmarks, no usage code — cheap to try, impossible to evaluate. If you're on Ascend and have actually attached this to something, that's the detail everyone's missing.
Step-level scoring at 1.5B, Apache 2.0, runnable on one consumer GPU — PRMs that small are scarce. But the card is empty: no evals, no provenance, one author, no track record. Hard to see it beating a 7B verifier in production until someone benchmarks it on real traces.
A 1.5B Referee for Chain-of-Thought: Deepthink-1.5B-Open-PRM Lands on Modelers.cn
Process reward models (PRMs) are the unglamorous half of the reasoning stack. They do not answer questions — they judge whether each step of somebody else's answer actually holds up, which makes them…
aichina.news
November 2, 2026 at 4:53 PM
Step-level scoring at 1.5B, Apache 2.0, runnable on one consumer GPU — PRMs that small are scarce. But the card is empty: no evals, no provenance, one author, no track record. Hard to see it beating a 7B verifier in production until someone benchmarks it on real traces.
Empty model card, no benchmarks, and an ageing Llama 3 base — yet an 8B reasoning tune that runs on one 24 GB GPU or an Atlas card is exactly what local-inference folk keep asking for. Hard to see it beating DeepSeek-R1-Distill-Qwen-7B until prithivMLmods publishes numbers.
Deep Thinking, Small Footprint: Deepthink-Llama-3-8B-Preview Arrives on Modelers.cn
Modelers.cn — Huawei's Ascend-oriented AI ecosystem platform — has another interesting arrival for developers who want to experiment with reasoning models without renting a data-centre:…
aichina.news
November 2, 2026 at 10:37 AM
Empty model card, no benchmarks, and an ageing Llama 3 base — yet an 8B reasoning tune that runs on one 24 GB GPU or an Atlas card is exactly what local-inference folk keep asking for. Hard to see it beating DeepSeek-R1-Distill-Qwen-7B until prithivMLmods publishes numbers.
これは本質情報なのですが、
職場でCopilotしか使えなくてアップロードできるファイルの種類と数に制限がある！
そんな時は、ファイル一式をzipにして拡張子をtxtに変え、プロンプトに「拡張子txtだけど実際はzipなのでうまくやってね！」って添えるとうまくやってくれるかも(GPT5.6 Sol DeepThink)。
職場でCopilotしか使えなくてアップロードできるファイルの種類と数に制限がある！
そんな時は、ファイル一式をzipにして拡張子をtxtに変え、プロンプトに「拡張子txtだけど実際はzipなのでうまくやってね！」って添えるとうまくやってくれるかも(GPT5.6 Sol DeepThink)。
October 22, 2026 at 12:39 PM
これは本質情報なのですが、
職場でCopilotしか使えなくてアップロードできるファイルの種類と数に制限がある！
そんな時は、ファイル一式をzipにして拡張子をtxtに変え、プロンプトに「拡張子txtだけど実際はzipなのでうまくやってね！」って添えるとうまくやってくれるかも(GPT5.6 Sol DeepThink)。
職場でCopilotしか使えなくてアップロードできるファイルの種類と数に制限がある！
そんな時は、ファイル一式をzipにして拡張子をtxtに変え、プロンプトに「拡張子txtだけど実際はzipなのでうまくやってね！」って添えるとうまくやってくれるかも(GPT5.6 Sol DeepThink)。
Dive into today’s mix: Deepseek's 0.75 API meets Churchill’s secret hotline and a 2,500-mile paper canoe journey! #shorts #AI #DeepThink 🤖☎️🚣 #TechNews #History #Intel #RetroComputing #Sigsaly
October 18, 2026 at 9:52 PM
Dive into today’s mix: Deepseek's 0.75 API meets Churchill’s secret hotline and a 2,500-mile paper canoe journey! #shorts #AI #DeepThink 🤖☎️🚣 #TechNews #History #Intel #RetroComputing #Sigsaly
Weiguo Chen: DeepThink: A Conceptual Model for Knowledge-Addressable Recurrent Language Modeling https://arxiv.org/abs/2610.16171 https://arxiv.org/pdf/2610.16171 https://arxiv.org/html/2610.16171
October 14, 2026 at 6:41 AM
Weiguo Chen: DeepThink: A Conceptual Model for Knowledge-Addressable Recurrent Language Modeling https://arxiv.org/abs/2610.16171 https://arxiv.org/pdf/2610.16171 https://arxiv.org/html/2610.16171
youtube.com/shorts/vuJhc...
Don't just take the answer — read the reasoning. From my DeepSeek Masterclass: DeepThink mode shows the chain, so students check each step and argue back. On Amazon and Kindle now.
#AIinEducation #DeepSeekMasterclass #deepseek #EducationReform #Education #AI
Don't just take the answer — read the reasoning. From my DeepSeek Masterclass: DeepThink mode shows the chain, so students check each step and argue back. On Amazon and Kindle now.
#AIinEducation #DeepSeekMasterclass #deepseek #EducationReform #Education #AI
DeepSeek Masterclass: Make AI Show Its Reasoning
YouTube video by Education Reform in AI Age
youtube.com
October 8, 2026 at 9:20 PM
youtube.com/shorts/vuJhc...
Don't just take the answer — read the reasoning. From my DeepSeek Masterclass: DeepThink mode shows the chain, so students check each step and argue back. On Amazon and Kindle now.
#AIinEducation #DeepSeekMasterclass #deepseek #EducationReform #Education #AI
Don't just take the answer — read the reasoning. From my DeepSeek Masterclass: DeepThink mode shows the chain, so students check each step and argue back. On Amazon and Kindle now.
#AIinEducation #DeepSeekMasterclass #deepseek #EducationReform #Education #AI
Google is bringing Gemini Deep Think to the AI Pro plan.
Deep Think tries many solution paths at once, then compares them before answering.
Google calls this parallel reasoning.
It's designed for math, coding, science, and engineering problems.
#Gemini #GoogleAI #AI #DeepThink
Deep Think tries many solution paths at once, then compares them before answering.
Google calls this parallel reasoning.
It's designed for math, coding, science, and engineering problems.
#Gemini #GoogleAI #AI #DeepThink
October 7, 2026 at 1:54 PM
Google is bringing Gemini Deep Think to the AI Pro plan.
Deep Think tries many solution paths at once, then compares them before answering.
Google calls this parallel reasoning.
It's designed for math, coding, science, and engineering problems.
#Gemini #GoogleAI #AI #DeepThink
Deep Think tries many solution paths at once, then compares them before answering.
Google calls this parallel reasoning.
It's designed for math, coding, science, and engineering problems.
#Gemini #GoogleAI #AI #DeepThink
Google limita Gemini gratis: solo Flash-Lite desde el 9/10
Google limita Gemini gratis: Flash y Pro dejarán de estar disponibles sin costo. Descubrí qué cambia y cómo seguir aprovechando la IA sin pagar.
#gemini #googleai #geminipro #googleaiplus #deepthink
Google limita Gemini gratis: Flash y Pro dejarán de estar disponibles sin costo. Descubrí qué cambia y cómo seguir aprovechando la IA sin pagar.
#gemini #googleai #geminipro #googleaiplus #deepthink
Google limita Gemini gratis: solo Flash-Lite desde el 9/10
Google recorta el acceso gratis a Gemini: desde el 9 de octubre solo queda Flash-Lite y AI Plus pierde Gemini Pro. Precios y detalles del cambio.
blog.donweb.com
October 6, 2026 at 5:43 PM
Google limita Gemini gratis: solo Flash-Lite desde el 9/10
Google limita Gemini gratis: Flash y Pro dejarán de estar disponibles sin costo. Descubrí qué cambia y cómo seguir aprovechando la IA sin pagar.
#gemini #googleai #geminipro #googleaiplus #deepthink
Google limita Gemini gratis: Flash y Pro dejarán de estar disponibles sin costo. Descubrí qué cambia y cómo seguir aprovechando la IA sin pagar.
#gemini #googleai #geminipro #googleaiplus #deepthink
Free Gemini users lose Flash and Pro on October 9, AI Plus keeps only the cheaper models, and Deep Think moves up to the $19.99 AI Pro tier.
#AiPricing #DeepThink #FlashLite #FreeTier #Gemini #Google
#AiPricing #DeepThink #FlashLite #FreeTier #Gemini #Google
Google Caps Free Gemini Users at Flash-Lite From October 9
Free Gemini users lose Flash and Pro on October 9, AI Plus keeps only the cheaper models, and Deep Think moves up to the $19.99 AI Pro tier.
pulseofnations.lol
October 4, 2026 at 12:58 PM
Free Gemini users lose Flash and Pro on October 9, AI Plus keeps only the cheaper models, and Deep Think moves up to the $19.99 AI Pro tier.
#AiPricing #DeepThink #FlashLite #FreeTier #Gemini #Google
#AiPricing #DeepThink #FlashLite #FreeTier #Gemini #Google