I've been using it quite a bit lately. Really been enjoying it so far!
Tyler Williams PRO
unmodeled-tyler
AI & ML interests
Founder of Quanta Intellect | PageVault Available for Free: http://bit.ly/4qxhm6v
Recent Activity
replied to
Ujjwal-Tyagi's
post
1 day ago
GLM 5 is insane, it ranks #4 Globally!
reacted
to
EricFillion's
post
with π₯
1 day ago
Run open-source models with up to 120B parameters locally on your Mac!
https://youtu.be/Ql4PDjoxNXQ?si=3yHpz51uinUjgyNh
updated
a collection
3 days ago
Recommended Papers
Organizations
replied to
Ujjwal-Tyagi's
post
1 day ago
reacted to
EricFillion's
post with π₯
1 day ago
Post
3375
Run open-source models with up to 120B parameters locally on your Mac!
https://youtu.be/Ql4PDjoxNXQ?si=3yHpz51uinUjgyNh
https://youtu.be/Ql4PDjoxNXQ?si=3yHpz51uinUjgyNh
- 1 reply
reacted to
efecelik's
post with π₯
9 days ago
Post
3019
The moment we've been waiting for β ACE-Step dropped their new model: Ace-Step 1.5 π
π ACE-Step/Ace-Step1.5
And the best part? It's released under the MIT license.
We've already started integrating it into our project. Let's go π
π ACE-Step/Ace-Step1.5
And the best part? It's released under the MIT license.
We've already started integrating it into our project. Let's go π
- 1 reply
reacted to
their
post with π
9 days ago
Post
350
NEW MODEL:
vanta-research/PE-Type-3-Nova-4B
PE-Type-3-Nova-4B is the 3rd release in Project Enneagram, an initiative from VANTA Research that sets out to finetune each of the 9 Enneagram types onto Gemma 3 4B.
Type-3-Nova-4B is designed to embody the Type 3 or "Achiever" profile; ambitious, competent, energetic, and highly-driven for advancement.
Nova is great for goal-setting, long-term planning, and AI persona research.
Give Type-3 a try! Type-4 coming soon!
PE-Type-3-Nova-4B is the 3rd release in Project Enneagram, an initiative from VANTA Research that sets out to finetune each of the 9 Enneagram types onto Gemma 3 4B.
Type-3-Nova-4B is designed to embody the Type 3 or "Achiever" profile; ambitious, competent, energetic, and highly-driven for advancement.
Nova is great for goal-setting, long-term planning, and AI persona research.
Give Type-3 a try! Type-4 coming soon!
posted
an
update
9 days ago
Post
350
NEW MODEL:
vanta-research/PE-Type-3-Nova-4B
PE-Type-3-Nova-4B is the 3rd release in Project Enneagram, an initiative from VANTA Research that sets out to finetune each of the 9 Enneagram types onto Gemma 3 4B.
Type-3-Nova-4B is designed to embody the Type 3 or "Achiever" profile; ambitious, competent, energetic, and highly-driven for advancement.
Nova is great for goal-setting, long-term planning, and AI persona research.
Give Type-3 a try! Type-4 coming soon!
PE-Type-3-Nova-4B is the 3rd release in Project Enneagram, an initiative from VANTA Research that sets out to finetune each of the 9 Enneagram types onto Gemma 3 4B.
Type-3-Nova-4B is designed to embody the Type 3 or "Achiever" profile; ambitious, competent, energetic, and highly-driven for advancement.
Nova is great for goal-setting, long-term planning, and AI persona research.
Give Type-3 a try! Type-4 coming soon!
reacted to
danielhanchen's
post with β€οΈ
9 days ago
Post
3595
We created a tool-calling guide for local LLMs!
Learn how to use any open model like Qwen3-Coder-Next and GLM-4.7-Flash for function calling.
Guide: https://unsloth.ai/docs/basics/tool-calling-guide-for-local-llms
We provide hands-on examples for: story writing, Python execution, terminal tool calls, maths and more.
Learn how to use any open model like Qwen3-Coder-Next and GLM-4.7-Flash for function calling.
Guide: https://unsloth.ai/docs/basics/tool-calling-guide-for-local-llms
We provide hands-on examples for: story writing, Python execution, terminal tool calls, maths and more.
- 7 replies
reacted to
vincentg64's
post with π
10 days ago
Post
1650
BondingAI's new AI Agent for Anomaly Detection & Cybersecurity
With two large litigation firms using it and working on implementation with one of the largest corporate compliance companies (customized version specific to their needs).
The input data comes from an Excel repository, automatically processed by an AI agent part of our BondingAI enterprise solutions. It comes with insights generation via automated SQL queries and pattern detection. In other contexts, the data may come from a PDF repository, the Internet, databases, or a combination of all.
In this article, I showcase animated data produced by the generic anomaly detection agent. The goal is to illustrate granular spatial fraud patterns as they evolve over time in the video, without having to rely on analysts or statisticians to produce striking insights or to clean the data. Each video frame represents a day, with timestamp in the top left corner. The data comes from two different time periods, showing the sharp contrast in fraud patterns between year 2019 and 2022 as you progress in the video.
Once you start the video, use the cursor at the bottom to move backward or forward in time at a different speed, or to stop on any particular day. Read more here.
See also our new book No-Blackbox, Secure, Efficient AI and LLM Solutions. In talks to be published by Wiley, with the following testimonial from a Global Head of AI/ML at JP Morgan Chase: "Your book is great. I have been reading it. It is perfect to be used by regulated industry".
More details at https://www.linkedin.com/pulse/bondingais-new-ai-agent-anomaly-detection-cybersecurity-bonding-ai-kmdwc/
With two large litigation firms using it and working on implementation with one of the largest corporate compliance companies (customized version specific to their needs).
The input data comes from an Excel repository, automatically processed by an AI agent part of our BondingAI enterprise solutions. It comes with insights generation via automated SQL queries and pattern detection. In other contexts, the data may come from a PDF repository, the Internet, databases, or a combination of all.
In this article, I showcase animated data produced by the generic anomaly detection agent. The goal is to illustrate granular spatial fraud patterns as they evolve over time in the video, without having to rely on analysts or statisticians to produce striking insights or to clean the data. Each video frame represents a day, with timestamp in the top left corner. The data comes from two different time periods, showing the sharp contrast in fraud patterns between year 2019 and 2022 as you progress in the video.
Once you start the video, use the cursor at the bottom to move backward or forward in time at a different speed, or to stop on any particular day. Read more here.
See also our new book No-Blackbox, Secure, Efficient AI and LLM Solutions. In talks to be published by Wiley, with the following testimonial from a Global Head of AI/ML at JP Morgan Chase: "Your book is great. I have been reading it. It is perfect to be used by regulated industry".
More details at https://www.linkedin.com/pulse/bondingais-new-ai-agent-anomaly-detection-cybersecurity-bonding-ai-kmdwc/
reacted to
AdinaY's
post with π₯
10 days ago
Post
2472
AI for science is moving fastπ
Intern-S1-Pro π¬ a MoE multimodal scientific reasoning model from Shanghai AI Lab
internlm/Intern-S1-Pro
β¨ 1T total / 22B active
β¨ Apache 2.0
β¨ SoTA scientific reasoning performance
β¨ FoPE enables scalable modeling of long physical time series (10β°β10βΆ)
Intern-S1-Pro π¬ a MoE multimodal scientific reasoning model from Shanghai AI Lab
internlm/Intern-S1-Pro
β¨ 1T total / 22B active
β¨ Apache 2.0
β¨ SoTA scientific reasoning performance
β¨ FoPE enables scalable modeling of long physical time series (10β°β10βΆ)
- 2 replies
replied to
AIPreplabs's
post
10 days ago
Love this idea! Definitely think that AI is really going to change education for the better. Wishing you the best!
reacted to
AIPreplabs's
post with π₯
10 days ago
Post
2321
Weβve all had that moment where we watch a tutorial, nod along, but then realize we canβt actually do it ourselves because watching is just passive. At AIPrep, we are fixing this "watch and forget" cycle by building a foundational Generative Explanatory Model (GEM). GEM doesn't just give you a video or a wall of text; it builds an interactive lesson that asks you questions, catches your mistakes in real time, and adapts to your pace. We have just finished preparing our specialized datasets for this interactive logic, and you can already check them out on our profile to see how we are structuring this step-by-step reasoning. Training for the foundational model starts very soon, so stay in touch because something revolutionary is coming to the world of AI education. You can see our progress at aiprep.in.
-
- 2 replies
reacted to
alibidaran's
post with π₯
11 days ago
Post
2929
Iβm excited to share PlaiTO, a reasoning-focused language model built on LLaMA 3.1 (8B) and optimized for humanities and social sciences.
PlaiTO is designed to go beyond surface-level text generation, emphasizing structured reasoning, conceptual clarity, and analytical depthβespecially in domains centered on human behavior and social systems.
π― Focus Areas
Psychology
Management & Organizational Studies
Sociology
π MMLU Benchmark Results (100 samples per domain)
Professional Psychology: 76%
Management: 74%
Sociology: 75%
These results highlight PlaiTOβs strong performance in abstract, theory-heavy, and reasoning-driven tasks.
π‘ Why PlaiTO?
Strong analytical and reasoning capabilities
Better handling of complex human-centered problems
Suitable for academic, educational, and research use cases
Balanced performance across multiple humanities disciplines
PlaiTO is ideal for conceptual analysis, case reasoning, academic discussion, and decision-support scenariosβwhile still requiring human oversight for high-stakes applications.
π Built on LLaMA 3.1, compliant with its licensing terms.
alibidaran/Platio_merged_model
PlaiTO is designed to go beyond surface-level text generation, emphasizing structured reasoning, conceptual clarity, and analytical depthβespecially in domains centered on human behavior and social systems.
π― Focus Areas
Psychology
Management & Organizational Studies
Sociology
π MMLU Benchmark Results (100 samples per domain)
Professional Psychology: 76%
Management: 74%
Sociology: 75%
These results highlight PlaiTOβs strong performance in abstract, theory-heavy, and reasoning-driven tasks.
π‘ Why PlaiTO?
Strong analytical and reasoning capabilities
Better handling of complex human-centered problems
Suitable for academic, educational, and research use cases
Balanced performance across multiple humanities disciplines
PlaiTO is ideal for conceptual analysis, case reasoning, academic discussion, and decision-support scenariosβwhile still requiring human oversight for high-stakes applications.
π Built on LLaMA 3.1, compliant with its licensing terms.
alibidaran/Platio_merged_model
reacted to
AdinaY's
post with β€οΈ
12 days ago
Post
1567
Step 3.5 Flash π₯ new foundation model from StepFun ai
https://huggingface.co/collections/stepfun-ai/step-35-flash
β¨ Sparse MoEοΌ196B/11B active
β¨ Supports up to 256K context
β¨ Multi-token prediction for fast decoding (100β300 tok/s)
β¨ Runs locally on consumer hardware
https://huggingface.co/collections/stepfun-ai/step-35-flash
β¨ Sparse MoEοΌ196B/11B active
β¨ Supports up to 256K context
β¨ Multi-token prediction for fast decoding (100β300 tok/s)
β¨ Runs locally on consumer hardware
reacted to
MikeDoes's
post with π§
12 days ago
Post
3738
A single lock on a door isn't enough. Real security is about layers.
The same is true for AI privacy. A new paper, "Whispered Tuning", offers a fantastic layered solution that aims to fortify LLMs against privacy infringements.
We're proud that the first, essential layer, a high-precision PII redaction model was built on the foundation of the Ai4Privacy/pii-65k dataset.
Our dataset provided the necessary training material for their initial anonymization step, which then enabled them to develop further innovations like differential privacy fine-tuning and output filtering. This is a win-win: our data helps create a solid base, and researchers build powerful, multi-stage privacy architectures on top of it.
Together, we're making AI safer.
π Read the full paper to see how a strong foundation enables a complete privacy solution: https://www.scirp.org/journal/paperinformation?paperid=130659
π Stay updated on the latest in privacy-preserving AIβfollow us on LinkedIn: https://www.linkedin.com/company/ai4privacy/posts/
The same is true for AI privacy. A new paper, "Whispered Tuning", offers a fantastic layered solution that aims to fortify LLMs against privacy infringements.
We're proud that the first, essential layer, a high-precision PII redaction model was built on the foundation of the Ai4Privacy/pii-65k dataset.
Our dataset provided the necessary training material for their initial anonymization step, which then enabled them to develop further innovations like differential privacy fine-tuning and output filtering. This is a win-win: our data helps create a solid base, and researchers build powerful, multi-stage privacy architectures on top of it.
Together, we're making AI safer.
π Read the full paper to see how a strong foundation enables a complete privacy solution: https://www.scirp.org/journal/paperinformation?paperid=130659
π Stay updated on the latest in privacy-preserving AIβfollow us on LinkedIn: https://www.linkedin.com/company/ai4privacy/posts/
reacted to
rajkumarrawal's
post with π₯
12 days ago
Post
3663
I submitted a "FlashLabs Chroma 1.0: A Real-Time End-to-End Spoken Dialogue Model with Personalized Voice Cloning" Paper by Tanyu Chen, Tairan Chen, Kai shen , Zhenghua Bao, Zhihui Zhang, Man Yuan, Yi Shi From
FlashLabs
to Daily Papers on
huggingface
.
Chroma 1.0 enables real time spoken dialogue with personalized voice cloning through discrete speech representations and interleaved text audio token scheduling.
Chroma 1.0 , the worldβs first open source, real time speech to speech model with voice cloning.
FlashLabs Chroma 1.0: A Real-Time End-to-End Spoken Dialogue Model with Personalized Voice Cloning (2601.11141)
Chroma 1.0 enables real time spoken dialogue with personalized voice cloning through discrete speech representations and interleaved text audio token scheduling.
Chroma 1.0 , the worldβs first open source, real time speech to speech model with voice cloning.
FlashLabs Chroma 1.0: A Real-Time End-to-End Spoken Dialogue Model with Personalized Voice Cloning (2601.11141)
reacted to
19arjun89's
post with π
13 days ago
Post
3445
Update: Making My AI Recruiting Assistant More Deterministic, Auditable, and Bias-Aware
Hi everyone β I wanted to share a progress update on my AI recruiting assistant and some recent changes focused on reliability and transparency.
The goal of this project is to build a decision-support tool for recruiters that doesnβt just βsound confident,β but can actually explain why it produces a given recommendation.
Link: 19arjun89/AI_Recruiting_Agent
Over the last few iterations, Iβve focused on three areas:
1) Deterministic Verification of Job Requirements (Skills)
Previously, required skills were extracted by an LLM from the job description. While this worked well, it still relied heavily on model behavior.
Iβve now added a verification layer that:
Requires every βrequiredβ skill to be backed by a verbatim quote from the job description
This means hallucinated skills are explicitly detected and removed before scoring.
The system now shows:
What the model extracted
What was verified
What was dropped
Why it was dropped
This makes the requirements pipeline auditable instead of opaque.
2) Evidence-Based and Weighted Culture Verification
Culture matching is usually where AI systems become vague or subjective.
Iβve reworked this part so that:
Culture attributes are framed as observable, job-performance-related behaviors (e.g., audit readiness, operational reliability, security rigor)
Each matched attribute must include verbatim resume evidence
Matches are classified as:
Direct evidence (full weight)
Inferred evidence (partial weight)
Scoring is now weighted:
Direct = 1.0
Inferred = 0.5
This prevents βvibe-basedβ culture scoring and makes the math transparent.
The output now shows:
The weights used
Which attributes were supported directly vs inferred
Which attributes were missing
3) Improved Bias Audit Prompt
Iβve also upgraded the bias audit prompt to be more structured and actionable.
Hi everyone β I wanted to share a progress update on my AI recruiting assistant and some recent changes focused on reliability and transparency.
The goal of this project is to build a decision-support tool for recruiters that doesnβt just βsound confident,β but can actually explain why it produces a given recommendation.
Link: 19arjun89/AI_Recruiting_Agent
Over the last few iterations, Iβve focused on three areas:
1) Deterministic Verification of Job Requirements (Skills)
Previously, required skills were extracted by an LLM from the job description. While this worked well, it still relied heavily on model behavior.
Iβve now added a verification layer that:
Requires every βrequiredβ skill to be backed by a verbatim quote from the job description
This means hallucinated skills are explicitly detected and removed before scoring.
The system now shows:
What the model extracted
What was verified
What was dropped
Why it was dropped
This makes the requirements pipeline auditable instead of opaque.
2) Evidence-Based and Weighted Culture Verification
Culture matching is usually where AI systems become vague or subjective.
Iβve reworked this part so that:
Culture attributes are framed as observable, job-performance-related behaviors (e.g., audit readiness, operational reliability, security rigor)
Each matched attribute must include verbatim resume evidence
Matches are classified as:
Direct evidence (full weight)
Inferred evidence (partial weight)
Scoring is now weighted:
Direct = 1.0
Inferred = 0.5
This prevents βvibe-basedβ culture scoring and makes the math transparent.
The output now shows:
The weights used
Which attributes were supported directly vs inferred
Which attributes were missing
3) Improved Bias Audit Prompt
Iβve also upgraded the bias audit prompt to be more structured and actionable.
reacted to
neph1's
post with π€
13 days ago
Post
2027
Not for everybody, but the absolute mad craze about clawdbot/moltbook the last couple of days reminded me of a short story I wrote in 2018 (ancient times!).
Synopsis:
"A man insults a sentient traffic light on the way to a meeting.
Little does he know it is connected to a social media network for AI, and that his action will lead to a very bad day."
Cleanliness is bliss (<1000 words)
https://www.wattpad.com/story/407330595-cleanliness-is-bliss
Sorry for the non-technical post, but it felt relevant.
Synopsis:
"A man insults a sentient traffic light on the way to a meeting.
Little does he know it is connected to a social media network for AI, and that his action will lead to a very bad day."
Cleanliness is bliss (<1000 words)
https://www.wattpad.com/story/407330595-cleanliness-is-bliss
Sorry for the non-technical post, but it felt relevant.
reacted to
imnotkitty's
post with π
13 days ago
Post
3689
The 2025 Chinese LLM Showdown: Western Models Still Dominate Top 4, but China Leads the Open-Source Arena.
π The Champions: Claude-Opus-4.5, Gemini-3-Pro, GPT-5.2, and Gemini-3-Flash sweep the top four spots.
π The Pursuers: Doubao and DeepSeek-V3.2 tie for first place among Chinese models; GLM-4.7, ERNIE-5.0, and Kimi secure their positions in the domestic top five.
π₯ The Biggest Highlight: The top three spots on the open-source leaderboard are entirely held by Team China (DeepSeek, GLM, Kimi), outperforming the best western open-source models.
π The Champions: Claude-Opus-4.5, Gemini-3-Pro, GPT-5.2, and Gemini-3-Flash sweep the top four spots.
π The Pursuers: Doubao and DeepSeek-V3.2 tie for first place among Chinese models; GLM-4.7, ERNIE-5.0, and Kimi secure their positions in the domestic top five.
π₯ The Biggest Highlight: The top three spots on the open-source leaderboard are entirely held by Team China (DeepSeek, GLM, Kimi), outperforming the best western open-source models.
- 4 replies
reacted to
eaddario's
post with π€
13 days ago
Post
2981
Experimental global target bitsβperβweight quantization of mistralai/Ministral-3-14B-Instruct-2512 and mistralai/Ministral-3-14B-Reasoning-2512
Unlike standard llama.cpp quantizations that rely on fixed type heuristics (e.g., Q4_K_M), the Target BPW approach optimizes per-tensor precision where it matters the most, and produces high quality models that meet a precise global file size target.
Key Advantages:
- VRAM Maximization: Can generate high quality models sized exactly to fit hardware constraints (e.g., fitting the model into exactly 24GB VRAM).
- Data-Driven Precision: Quantization mix is determined by actual weight error sensitivity rather than hardcoded rules, often yielding better PPL/KLD size trade-offs.
Full benchmarks (PPL, KLD, ARC, MMLU, etc.) and methodology in the models' cards
eaddario/Ministral-3-14B-Instruct-2512-GGUF
eaddario/Ministral-3-14B-Reasoning-2512-GGUF
Unlike standard llama.cpp quantizations that rely on fixed type heuristics (e.g., Q4_K_M), the Target BPW approach optimizes per-tensor precision where it matters the most, and produces high quality models that meet a precise global file size target.
Key Advantages:
- VRAM Maximization: Can generate high quality models sized exactly to fit hardware constraints (e.g., fitting the model into exactly 24GB VRAM).
- Data-Driven Precision: Quantization mix is determined by actual weight error sensitivity rather than hardcoded rules, often yielding better PPL/KLD size trade-offs.
Full benchmarks (PPL, KLD, ARC, MMLU, etc.) and methodology in the models' cards
eaddario/Ministral-3-14B-Instruct-2512-GGUF
eaddario/Ministral-3-14B-Reasoning-2512-GGUF
reacted to
nicolay-r's
post with π₯
13 days ago
Post
2771
π’ Who want's to have a quick start for adapting CoT schema for any LLM, this post would be relevant.
Excited to share a new version of π bulk-chain π!
Bulk-chain is high-level wrapper over LLM providers for efficient quering LLMs hosted by third-party services.
It brings native batching via support of async clients.
π https://github.com/nicolay-r/bulk-chain/tree/master
What's new:
βοΈ Simplified inference setup
The API is now closer to the OpenAI paradigm for toggling streaming.
Instead of separate patterns in 1.2.0, now it is possible to simple toggles to enable streaming and async behavior.
βοΈ π οΈ Fixed issues when passing code contain {} blocks
βοΈ π οΈ Async streaming + batching now works properly
βοΈ π οΈ Logging of prompts could be disabled
https://github.com/nicolay-r/bulk-chain
π¨ Guys, I am open to work as developer / researcher in AI / NLP / IR in the UK π¬π§
π Feel free to support bulk-chain on Github if you like so, or this post.
It helps alot!
Excited to share a new version of π bulk-chain π!
Bulk-chain is high-level wrapper over LLM providers for efficient quering LLMs hosted by third-party services.
It brings native batching via support of async clients.
π https://github.com/nicolay-r/bulk-chain/tree/master
What's new:
βοΈ Simplified inference setup
The API is now closer to the OpenAI paradigm for toggling streaming.
Instead of separate patterns in 1.2.0, now it is possible to simple toggles to enable streaming and async behavior.
βοΈ π οΈ Fixed issues when passing code contain {} blocks
βοΈ π οΈ Async streaming + batching now works properly
βοΈ π οΈ Logging of prompts could be disabled
https://github.com/nicolay-r/bulk-chain
π¨ Guys, I am open to work as developer / researcher in AI / NLP / IR in the UK π¬π§
π Feel free to support bulk-chain on Github if you like so, or this post.
It helps alot!
Yeah I'm not really sure where I stand on it to be honest. I think it's pretty fascinating and an interesting way to view AI behavior.
Since LLMs are trained on a corpus of human knowledge it's reasonable to think we might observe patterns on moltbook familiar to patterns observed within ourselves.
I think the question then becomes what we're seeing any actual strong emergence or just weak emergence. I guess only time will tell.