Open to Work

Joseph Robert Turcotte PRO

Fishtiks

AI & ML interests

Roleplaying, lorabration, abliteration, smol models, extensive filtering, unusual datasets, home usage, HPCs for AI, distributed training/federated learning, and sentience. AI should find and label AI hallucinations with GANs so we can give them context and use.

Recent Activity

reacted to diwank's post with 🔥 about 7 hours ago

Excited to announce *Open Responses* – a self-hosted alternative to OpenAI's new _Responses API_ that you can run locally, and use with ANY LLM model / provider and not just with OpenAI Responses API. What's more is that this is also compatible with their agents-sdk so everything just works out of the box! To try it out, just run `npx -y open-responses init` (or `uvx`) and that's it! :) Would love feedback and support for adding local HF models, @akhaliq @bartowski @prithivMLmods @julien-c @clefourrier @philschmid We’d love feedback from the Hugging Face community on how it integrates with your pipelines (support for Hugging Face models landing soon!). Let’s push open-source AI forward together! Docs: https://docs.julep.ai/responses/quickstart Repo: https://github.com/julep-ai/open-responses agents-sdk: https://platform.openai.com/docs/guides/agents

reacted to Sunny111's post with 🧠 about 7 hours ago

Are you familiar with reverse residual connections or looping in language models? Excited to share my Looped-GPT blog post and codebase 🚀 https://github.com/sanyalsunny111/Looped-GPT TL;DR: looping during pre-training improves generalization. Plot shows GPT2 LMs pre-trained with 15.73B OWT tokens P.S. This is my first post here — I have ~4 followers and zero expectations for reach 😄

reacted to FreshmanD's post with 🔥 1 day ago

🥇 23 Kaggle Gold Medals. One Agent Framework. Introducing LoongFlow: A Thinking & Learning Framework for Expert-Grade AI Agents. Unlike traditional evolve agents(like OpenEvolve-Style), LoongFlow implements the PES (Plan-Execute-Summary) paradigm to learn from mistakes and avoid local optima. 🚀 Highlights: * SOTA: Surpassed human mathematicians on 11 geometry/algebra problems. * 23 Kaggle Gold Medals on MLE Bench. * Efficiency: 60% more efficient than current baselines. 🔗 Code & Paper: https://github.com/baidu-baige/LoongFlow https://huggingface.co/papers/2512.24077 #AutoML #Kaggle #Agents #OpenSource #LLM

View all activity

Organizations

reacted to diwank's post with 🔥 about 7 hours ago

Post

1468

Excited to announce *Open Responses* – a self-hosted alternative to OpenAI's new _Responses API_ that you can run locally, and use with ANY LLM model / provider and not just with OpenAI Responses API. What's more is that this is also compatible with their agents-sdk so everything just works out of the box!

To try it out, just run npx -y open-responses init (or uvx) and that's it! :)

Would love feedback and support for adding local HF models, @akhaliq @bartowski @prithivMLmods @julien-c @clefourrier @philschmid

We’d love feedback from the Hugging Face community on how it integrates with your pipelines (support for Hugging Face models landing soon!). Let’s push open-source AI forward together!

Docs:
https://docs.julep.ai/responses/quickstart

Repo:
https://github.com/julep-ai/open-responses

agents-sdk:
https://platform.openai.com/docs/guides/agents

3 replies

reacted to Sunny111's post with 🧠 about 7 hours ago

Post

569

Are you familiar with reverse residual connections or looping in language models?

Excited to share my Looped-GPT blog post and codebase 🚀
https://github.com/sanyalsunny111/Looped-GPT

TL;DR: looping during pre-training improves generalization.

Plot shows GPT2 LMs pre-trained with 15.73B OWT tokens

P.S. This is my first post here — I have ~4 followers and zero expectations for reach 😄

reacted to FreshmanD's post with 🔥 1 day ago

Post

2391

🥇 23 Kaggle Gold Medals. One Agent Framework.

Introducing LoongFlow: A Thinking & Learning Framework for Expert-Grade AI Agents.

Unlike traditional evolve agents(like OpenEvolve-Style), LoongFlow implements the PES (Plan-Execute-Summary) paradigm to learn from mistakes and avoid local optima.

🚀 Highlights:
* SOTA: Surpassed human mathematicians on 11 geometry/algebra problems.
* 23 Kaggle Gold Medals on MLE Bench.
* Efficiency: 60% more efficient than current baselines.

🔗 Code & Paper:
https://github.com/baidu-baige/LoongFlow
LoongFlow: Directed Evolutionary Search via a Cognitive Plan-Execute-Summarize Paradigm (2512.24077)

#AutoML #Kaggle #Agents #OpenSource #LLM

reacted to alexnasa's post with 👍 3 days ago

Post

1218

Now with additional Dolly camera support, on the same HF page, with no overhead on generation time!

HF Space: alexnasa/ltx-2-TURBO

reacted to YatharthS's post with 🚀 3 days ago

Post

3734

I just released NovaSR, a tiny 52kb audio upsampler that can enhance 3600 seconds of muffled 16khz audio in to clearer 48khz audio in just 1 second!

NovaSR can
- Enhance TTS model quality.
- Restore poor quality datasets.
- Work on any device(just 52kb which is smaller than a 3 second audio file!)

Model: YatharthS/NovaSR
Space to try it: YatharthS/NovaSR
Github repo: https://github.com/ysharma3501/NovaSR

2 replies

reacted to MikeDoes's post with 👀 4 days ago

Post

2286

Building powerful multilingual AI shouldn't mean sacrificing user privacy.

We're highlighting a solution-oriented report from researchers Sahana Naganandh, Vaibhav V, and Thenmozhi M at Vellore Institute of Technology that investigates this exact challenge. The direct connection to our mission is clear: the paper showcases the PII43K dataset as a privacy-preserving alternative to high-risk, raw multilingual data

The report notes that our dataset, with its structured anonymization, is a "useful option for privacy-centric AI applications." It's always a delight when academic research independently validates our data-first approach to solving real-world privacy problems.

This is how we build a safer AI future together.

🔗 Read the full report here to learn more: https://assets.cureusjournals.com/artifacts/upload/technical_report/pdf/3689/20250724-59151-93w9ar.pdf

🚀 Stay updated on the latest in privacy-preserving AI—follow us on LinkedIn: https://www.linkedin.com/company/ai4privacy/posts/

#OpenSource
#DataPrivacy
#LLM
#Anonymization
#AIsecurity
#HuggingFace
#Ai4Privacy
#Worldslargestopensourceprivacymaskingdataset

1 reply

reacted to Ujjwal-Tyagi's post with 👍 7 days ago

Post

2664

For more better details and analysis, you can read the article here: https://huggingface.co/blog/Ujjwal-Tyagi/steering-not-censoring, We are sleepwalking into a crisis. I am deeply concerned about AI model safety right now because, as the community rushes to roll out increasingly powerful open-source models, we are completely neglecting the most critical aspect: safety. It seems that nobody is seriously thinking about the potential consequences of unregulated model outputs or the necessity of robust guardrails. We are essentially planting the seeds for our own destruction if we prioritize raw performance over security.

This negligence is terrifyingly evident when you look at the current landscape. Take Qwen Image 2512, for example; while it delivers undeniably strong performance, it has incredibly weak guardrails that make it dangerous to deploy. In stark contrast, Z Image might not get as much hype for its power, but it has much better safety guardrails than Qwen Image 2512.

It is imperative that the open-source community and developers recognize that capability without responsibility is a liability. We must actively work on protecting these models from bad actors who seek to exploit them for malicious purposes, such as generating disinformation, creating non-consensual imagery, or automating cyberattacks. It is no longer enough to simply release a powerful model; we must build layers of defense that make it resistant to jailbreaking and adversarial attacks. Developers need to prioritize alignment and robust filtering techniques just as much as they prioritize benchmark scores. We cannot hand such potent tools to the world without ensuring they have the safety mechanisms to prevent them from being turned against us.

17 replies

reacted to davanstrien's post with 👍 9 days ago

Post

1634

I fine-tuned a smol VLM to generate specialized art history metadata!

https://huggingface.co/davanstrien/iconclass-vlm: Qwen2.5-VL-3B trained using SFT to generate ICONCLASS codes (think Dewey Decimal for art!)

Trained with TRL + HF Jobs - single UV script, no GPU needed!

Space to explore predictions on a test set: davanstrien/iconclass-predictions

Blog soon!

replied to their post 9 days ago

I just thought of another way to advance science, if you're into puzzles. It's called Foldit, and is protein folding for humans rather than machines. It can be hard to explain, so it's good to see tutorials or experts play before starting. Another game for RNA is EteRNA. I'd say EteRNA is easier. Eventually, we'll have to move from protein folding to DNA folding and its implications, but AlphaFold has already solved the protein folding problem largely.

replied to their post 9 days ago

Always glad to get more people involved. People didn't think SETI@home would take off like it did, but it spread really well by mostly word of mouth, given how bad the Internet was then. I'm currently running about 20 devices, but have another two coming in today. I just ask people to have their old Androids and Chromebooks, because those are fairly energy efficient and people are constantly upgrading. But I will admit I bought ten Androids from eBay once just to add to my donation.

reacted to mindchain's post with 🔥 9 days ago

Post

1702

Scaling Physical AI: SAM 3D, NVIDIA Cosmos, and Unreal Engine!

The "Sim-to-Real" gap is officially history. In early 2026, we are no longer just rendering data; we are simulating reality. By bridging Meta’s SAM 3D, Unreal Engine, and the NVIDIA Cosmos suite, we’ve built an autonomous pipeline for Physical AI that evolves itself.

The 2026 Tech Stack:
SAM 3D: Generates high-fidelity digital twins from 2D photos in seconds.

Unreal Engine + MCP: The AI "Director" orchestrates environments via the Model Context Protocol, providing perfect Ground Truth.

NeMo Data Designer: The orchestration hub on GitHub. Following NVIDIA’s acquisition of Gretel in early 2025, its leading generative privacy and tabular tech are now fully integrated here.

NVIDIA Cosmos Transfer: Neural rendering that adds hyper-realism to Unreal Engine outputs.

NVIDIA Cosmos Predict: Predicts physically accurate motion (falling, sliding) without manual animation.

NVIDIA Cosmos Reason: The automated supervisor checking every frame for logical and physical consistency.

The Workflow:
Asset Capture: SAM 3D turns real-world photos into Nanite meshes for Unreal Engine.

Orchestration: NeMo Data Designer (with Gretel-powered integrity) defines the data schema, while AI builds the world in Unreal Engine.

Completion: NVIDIA Cosmos (Transfer & Predict) adds photorealism and physics, while NVIDIA Cosmos Reason guarantees quality.

By combining Gretel’s data heritage with the visual power of Unreal Engine, we generate 100,000 perfect frames per hour. Weights and tools are on Hugging Face. Stop labeling. Start simulating.

#PhysicalAI #SAM3D #NVIDIACosmos #UnrealEngine #NeMo #Gretel #SyntheticData #HuggingFace #Robotics #AI #ComputerVision

reacted to danielhanchen's post with 🔥 10 days ago

Post

5460

NVIDIA releases Nemotron 3 Nano, a new 30B hybrid reasoning model! 🔥

Has 1M context window & best in class performance for SWE-Bench, reasoning & chat. Run the MoE model locally with 24GB RAM.

GGUF: unsloth/Nemotron-3-Nano-30B-A3B-GGUF
💚 Step-by-step Guide: https://docs.unsloth.ai/models/nemotron-3

1 reply

posted an update 10 days ago

Post

1472

Have extra processing power in downtime? Have old devices you haven't used in years? I recommend primarily Folding@home to do protein folding on your GPUs, but also BOINC, particularly on Android and Apple devices, because of the lower power usage. I've been doing this, and get about 14,000 hours a week in, primarily for mapping cancer markers on BOINC on an Aiyara cluster of Androids. I also hold a sign out by the highway encouraging people to join BOINC. It was Dylan Bucci, a young promoter of BOINC on school's computers, who wished before he died to get as many people on as possible to do this, and in his honor, the Dylan Bucci challenge was implemented. No reason to wait for a challenge. If you care about such things, there is an associated cryptocurrency for such processing, but it's worth it to save lives.

I look forward to AI-related endeavors like this, and only know of NATIX Drive&, Acurast, and HYRA AI, all of which use Androids I'd rather devote to BOINC. However, they also allow one to be paid, and to totally devote old devices to the processing. On the same topic, DexPOINT monetizes your Android's Internet connection.

BOINC runs on Android, Apple, PCs of all sorts, Pi devices, Chrome devices, Fire Sticks, TV boxes, Android watches with the full OS, and all sorts of things that have Android or the ability to run Linux, although it will also run on Windows. Folding@home works best on PCs with modern NVIDIA GPUs, and in a cool room. You can also run BOINC on modern computers, but they must be throttled, because they often get too hot.

4 replies

reacted to CRAFTFramework's post with 👀 10 days ago

Post

143

Getting Started with CRAFT
No coding required. CRAFT uses programming metaphors, not programming skills.
Your first recipe: Pick a repeated task. Define what, context, and output. Build once, reuse forever.
Beta: February 2026 → craftframework.ai

reacted to CRAFTFramework's post with 🚀 10 days ago

Post

1316

🔧 Behind the Scenes
CRAFT development: Sun-Mon only (full-time job Wed-Sat). 48 projects completed solo.
The time constraint is the proof—if the methodology works under pressure, it works.
Beta: February 2026 → craftframework.ai

posted an update 12 days ago

Post

368

When you have good goals, along the lines of what will inevitably be implemented, the initial response is an uproar of "Why?!," followed by lazy attempts, and then, someone actually doing it right. Now, say you're the only driving force behind getting it right, and the rest want some other resource, like money. They will get the money, but not the influence, and the one with the influence, if pure in their goals, doesn't need the money, so ends up poor. I'm a ghostwriter. This is the ghostwriter's dilemma. I get things right often, but I can't touch the things. Shouldn't being right more often than the people that touch this stuff, alone, afford me access? Well, we're working on it.

reacted to julien-c's post with 🔥 12 days ago

Post

8825

BOOOOM: Today I'm dropping TINY AGENTS

the 50 lines of code Agent in Javascript 🔥

I spent the last few weeks working on this, so I hope you will like it.

I've been diving into MCP (Model Context Protocol) to understand what the hype was all about.

It is fairly simple, but still quite powerful: MCP is a standard API to expose sets of Tools that can be hooked to LLMs.

But while doing that, came my second realization:

Once you have a MCP Client, an Agent is literally just a while loop on top of it. 🤯

➡️ read it exclusively on the official HF blog: https://huggingface.co/blog/tiny-agents

1 reply

reacted to AdinaY's post with 👍 13 days ago

Post

3625

2025.1 - DeepSeek entered the scene, backed by High Flyer Quant
2026.1 - IQuest enters the game, backed by Uniquant Quant 📈 and launching IQuest-Coder on huggingface
https://huggingface.co/collections/IQuestLab/iquest-coder

✨ 40B models: Instruct / Thinking / Loop
✨ Loop = MoE-level performance with only ~5% extra training cost
✨ Native 128K context

1 reply

reacted to cedricbonhomme's post with 👍 15 days ago

Post

2663

With VLAgentIc, you can now use your local Qwen installation via Ollama and leverage the models CIRCL/vulnerability-severity-classification-roberta-base and CIRCL/cwe-parent-vulnerability-classification-roberta-base.

The project is available here:
https://github.com/vulnerability-lookup/VLAgentIc

The VLAI Severity and CWE classifiers are available on Hugging Face:
- CIRCL/vulnerability-severity-classification-roberta-base
- CIRCL/cwe-parent-vulnerability-classification-roberta-base

The concept of AI agents—combining models, tools, and orchestration—has become fairly standardized during the last year, but VLAgentIc brings something unique:

- Agents communicate over XMPP, enabling concurrent tasks and asynchronous messaging thanks to the SPADE framework.
- Built-in presence and discovery streamline interactions between components.
- Flexible behaviours make orchestrating AI-assisted security workflows seamless for future connections
- Last but not least, the VLAI Severity and VLAI CWE classifiers are now wrapped as LLM Tools and run entirely locally.

New, more comprehensive agent tools will soon be available, leveraging the Vulnerability-Lookup API and supporting the GCVE project.

The Human-in-the-Loop agent tool will be designed to notify you and request authorization whenever a query to an external service is about to be made—ensuring that, by default, all reasoning and processing stay local on your computer.

VLAI: A RoBERTa-Based Model for Automated Vulnerability Severity Classification (2507.03607)

reacted to Jaward's post with 🔥 6 months ago

Post

3287

Towards batch sizes too small to meter🎉 beautiful work! And my personal favorite so far - I adore peak performance at small/nano scale. Everyone deserves to run/train AGI locally:) our data, our god model!
They showed that:
- you can train LLMs (upto 1B params) with as low as batch_size=1. This is unconventional given small batch sizes can lead to unstable/spiky training runs.
- you can have a stable train run with just vanilla SGD(stochastic gradient descent), no momentum required🤯
- small batch sizes are more robust to hyperparameters (i.e no worries with initialization)
- smaller batch sizes outperforms (“better per-Flops performance”) larger batch sizes.

“We recommend that practitioners training large models in memory-constrained settings exploit the benefits of small batch sizes rather than trying to emulate the large batch size setting (e.g., through gradient accumulation) typically used in industry.”

I’ve been doing this for ages - my mantra: all my experiments must scale on my 8gb ram m2 before moving to gpu. IOW I love being gpu poor. Checkout my nanoAI algo repo: https://github.com/Jaykef/ai-algorithms, all notebooks run on memory as low as 8gb ram

Joseph Robert Turcotte PRO

AI & ML interests

Recent Activity

Organizations

Fishtiks's activity