Archad Sawadogo

[AmirMušić @AmirMushich]

GPT-6 Astra (or Sol) / Opus 5.5. + my SKILL.md:
turn real estate photos into 3D walkable spaces

the Skill will guide you - save it:
https://t.co/MKQeBG9Iy7

[Ryan Sael @RyanSael]

this shouldn't be possible.

publishing all 3 soon!

[Christine Park @sparkcpark]

Introducing Synthetic Hospital: an open, fully synthetic longitudinal EHR benchmark with verifiable ground truth!

1,268 patients, 5,602 encounters, zero PHI. Physicians could not reliably distinguish its charts from real ones.

📄 https://t.co/kU6Rkqr2dA
💻 https://t.co/k3UH8vgQ6H
✍️ https://t.co/otOqD6LMee

[Tim Dettmers @Tim_Dettmers]

We built an all-synthetic simulation of a hospital that can be shared publicly without any issues. Our data quality is so high, physicians cannot tell the difference between real/synthetic patients. The data is fully verified -- perfect for RL.

[DON “Tobechukwu” Ade 👑 @Row_Haastrup]

I’ve been using Amazon for over 8 years and I am just discovering this 😭 😭😭😭😭😭😭😭😭😭😭😭😭😭😭😭😭

[Dex @DexxterUtd]

the origin of manchester city's cheating but nobody said a word because they were all desperate to dethrone united

[妖精アーヤ @aiehon_aya]

Claude Opus5.5 モーショングラフィックス
────────────────────
天才Opus5.5にどうやって作ったのか
説明させてみた（↓）
─────────
■ 全体の流れ
index.html（絵を描くプログラム）
↓ 1コマずつ描かせる
render .py（900コマを撮って動画ファイルにする）
↓
make_audio.py（音楽を計算で作る）
↓
合体して MP4

① すべてを「拍」で決めている
いちばん大事な考え方です。テンポを 128BPM にすると、32拍でちょうど15秒になります。「4拍目で MOVE が出る」「28拍目で爆発」のように、映像の出来事をすべて拍の番号で書いています。音楽も同じ拍の番号で鳴らしています。だから、映像と音がずれる余地がありません。手作業のタイミング合わせは一度もしていません

② 動きは「イージング（緩急）」の数式
たとえば文字が滑り込む動きは、「0→1 に進む割合」を数式で曲げて作っています。最初速く、最後ゆっくり止まる（expo）：MOVE の滑り込み
少し行き過ぎて戻る（back）：SHAPE の文字がポンと出る
跳ねて止まる（bounce）：RHYTHM の文字が落ちて弾む

グラフエディタの場面はこの「緩急の曲線」そのものを見せる演出です。モーションデザイナーが毎日さわる道具なので分かる人ほど「おっ」となるところです。

③ 図形の変形・3D・粒子もすべて計算
丸→四角→三角→星：図形のまわりを240点で表し、同じ点同士を少しずつ動かして変形させています
トンネルとワイヤーの立方体：遠くの物ほど小さく描く計算（遠近法）を自分で書いて、3Dっぽく見せています。

最後の CLAUDE：一度見えないところに文字を描き、文字の形の点を約5,000個拾っています。その位置へ粒子を集めています。

④ 仕上げの質感
モーションブラー：1コマを、ほんの少しずつ時間をずらして10回描き、平均しています。速い動きがなめらかにぶれて、映像らしくなります（5回だと残像が段々に見えたので10回にしました）
グリッチ：赤・緑・青の色を分けて左右にずらし、画面を横に切ってずらしています
フィルムの粒子・周辺の暗さ：薄いノイズを毎コマ違う位置に重ねています
四隅の表示（タイ��コードなど）：「白黒反転」の重ね方で描いているので、どの背景色でも読めます

⑤ 書き出し
render. py が、画面を持たない Chrome を裏で起動します。「0コマ目を描いて → 画像を受け取る」を900回くり返し、そのまま ffmpeg（動画を作る道具）へ流して MP4 にしています。約6分かかりました。

⑥ 音楽も計算で作っている
楽器の音を、波の形の計算で作っています。
キック：低い音が一瞬で下がっていく波
ハット・スネア・「シュッ」という効果音：ザーッという雑音を削って形を整えたもの
和音：少しずつ音程をずらしたノコギリ波を7本重ねたもの

さらに、キックが鳴るたびに他の音を一瞬小さくして、クラブ音楽っぽいうねりを出しています。最後の決めの直前には、わざと0.1秒ほど完全な無音を入れて、爆発の落差を大きくしています。

[TiredAndOnFire-HD Remasters/Gamedev @TIREDnONFIRE]

Monster Hunter Portable 3rd has been recompiled for windows PC.
Complete with menus for config adjustments, online multiplayer support, and drag and drop HD texture packs.

Can't wait to try this one!

[Ripper's Archive @rippersarchive]

Blood: The Last Vampire (2000)

1080p (5.69GB)

Audio: JA
Subs: EN

&
the spin-off anime

Blood+ (2005-2006)

All Fifty Episodes

1080p (32.68GB)

Audio: EN, JA
Subs: EN

&
the other spin-off anime

Blood-C (2011)

All Twelve Episodes

1080p (22.1GB)

Audio: EN, JA
Subs: EN

[Ripper's Delight @rippersdelight]

Mobile Suit Gundam: The 08th MS Team (1996-1999)

All Twelve Episodes

1080p Remux (84.66GB)

Audio: English, Japanese
Subtitles: English

[Innopro @innopro_]

THIS IS FUCKING INSANE.

a Chinese developer just released a FREE tool that can turn one script and one photo into a complete video with a presenter.

it was just released and already has 26,000 stars.

it’s called
"lanshu-create-ai-presenter-video".

here's how it works:

you give it a script and a presenter photo you have permission to use.

the AI creates the voice, makes the presenter talk, and matches the lips to the audio.

then it adds the subtitles, creates the cover, checks that everything is synced properly, and gives you the finished video with a quality report.

the final video is 9:16, 1080x1920, 30fps, and 45-75 seconds long.

and the crazy part is

it does all of this in one run on your own computer.

if you make short form content, save this before you make your next video.

will be dropping more free tools soon,
so bookmark

[Akshay 🚀 @akshay_pachaar]

NVIDIA and Stanford just challenged Jev.

(their new System 1 architecture runs up to 9x faster.)

It is called a Contrastive Language Model, or CLM.

Like Jev, CLM is not designed to generate text. It handles the small, repeated decisions inside AI systems, such as choosing a tool, ranking a patch, routing a request, or selecting the next action.

But CLM reaches those decisions differently.

Instead of generating an answer token by token, it treats decision-making as a retrieval problem.

Here is how it works.

1) Encode the state

CLM takes the current situation, such as an agent’s context or the state of a game, and converts it into a vector.

It uses a frozen Qwen3-8B model with a small trainable state projection head.

2) Encode every possible action

A separate action head converts each candidate into the same vector space.

In the Mario example, the candidates are left, jump, and right run. CLM does not invent a fourth option. It only evaluates the actions supplied by the application.

3) Learn which states and actions belong together

During training, the correct state-action pair is pulled closer while incorrect pairs are pushed apart.

A batch of B examples produces a B × B similarity matrix. The matching pairs sit on the diagonal. Every other pairing becomes a negative example.

This contrastive training uses InfoNCE, the same general mechanism behind systems such as CLIP and dense retrieval.

4) Turn similarity into a decision

At inference, CLM measures the cosine similarity between the state and every candidate action.

A softmax converts those scores into a probability distribution. The application can choose the winner, apply a confidence threshold, or escalate an uncertain result.

The real speed advantage comes from separating states and actions.

Actions can be embedded once and cached. If an agent repeatedly chooses between the same tools, CLM only needs to encode the changing state and compare it with stored action vectors.

That replaces repeated generation with one embedding pass and a set of cheap dot products.

The researchers report that CLM-8B matches Jev across computer-use, gaming, and tool-calling evaluations while reaching up to 9x lower latency. The improvement is largest when actions repeat or the candidate set grows.

CLM still has limits. It cannot generate new actions, its probabilities are relative to the supplied candidates, and its strongest verifier results require task-specific fine-tuning.

But its central idea is powerful.

The entire research is open-source, including the code.

Read more here: https://t.co/I9kPwMPI7B

When software already knows the possible answers, an AI model should score them instead of generating more words.

I also wrote a full breakdown on how system one models like Jev work.

The article is quoted below.

[NVIDIA Healthcare @NVIDIAHealth]

Introducing MONAI Physio at #MICCAI2026.

3D and 4D medical images ➡️ personalized cardiac and respiratory digital twins.

This open-source Project MONAI toolkit lets researchers model how the heart beats and lungs move for simulation, visualization, and reproducible research.

Explore the project 👉 https://t.co/jmFhZW5k4R

[Bad Decisions Studio @badxstudio]

Blender + GPT-6 Astra + Unreal Engine 🚀

[Meng To @MengTo]

I recorded a 47-minute tutorial on how I use Opus 5.5 to create designs with gorgeous three.js scenes.

I've been using Mobbin MCP to give my agents strong references, then letting Opus bring them together into a full landing page with multiple sections and a consistent design.

Opus also created a super detailed brand guide, logo explorations, and some amazing ad creatives. Went way beyond my expectations.

Live site: https://t.co/XvMDmU4EAI

Site made during the tutorial, with prompts: https://t.co/5edYikmZEG

[AmirMušić @AmirMushich]

GPT Astra / Opus 5.5 -> Real estate photos to 3D walkable spaces - GitHub Skill

Repo is here:
https://t.co/1duqWGKmhJ

This workflow sparked a lot of "how to do it" questions - so we decided that to share the skill will be the best solution

Your AI agent will lead you through all the steps, and the README file will explain all the installation details.

Have fun:

[Hugging Apps @HuggingApps]

medical ai keeps advancing, now you can take the bones out of a chest X-ray 🩻

@qure_ai released a model that splits a radiograph into soft tissue, bone and lungs, so the ribs stop hiding what's behind them

▶️ on Spaces https://t.co/iQJ4651mmN

[金尘马 @jinchenma_ai]

我建议大家除了多刷X、Youtube看一手信息，优质论文也要多看，在 AI 辅助下，看这些内容已经没什么门槛了。

我一个非科研狗，用 ScienceBuddy 也能轻松看论文了。

现在我经常凌晨两三点，甚至四点才睡觉，早上十点、十一点才起床。

家里人也多次让我早点睡，我自己其实也想。但试过好几次，一想早睡就睡不着，然后又得折腾到天亮，反而睡得更不好。

于是我就想着，关于睡眠，有没有什么科学的做法，能帮我解决这个问题？

最近 Anthropic 也分享了 Claude 优化生物领域模型代码的进展，AI 已经开始参与一些具体的科研工作了。

ScienceBuddy 背后的团队 @PhAILabs 也开源了 ScienceIDE，把多个领域的真实科研代码整理成任务，让 AI 练习修复代码，再检查算出来的结果能不能对得上。

正好这次受邀尝试 PhAI Labs 的 ScienceBuddy，一个目前免费的 AI 科研工作台，我就想拿自己的睡眠问题试试，看看它能不能帮我解读相关论文。

于是我找了三份睡眠相关的论文和指南，上传到了 ScienceBuddy，让它帮我解读，再把里面的研究结果、适用条件和原文出处整理成一张表。看完有不明白的，就继续追着接着问，让它补查一些研究内容。

它整理出来的结果里提到，晚睡不一定就等于不健康，还得看睡眠��长够不够、作息是不是规律。

哎，看完结果，我感觉心情好多了，至少缓解了一部分焦虑。因为我不用光盯着自己几点睡，还可以先想办法把睡眠时间保证好。

另一个让我印象比较深的，就是它整理出来的调整思路：想调整作息，得先保证足够的总睡眠。

我现在是自由职业，时间还算自由，晚睡了可以安排晚一点起，睡够了，白天才有精神头。我准备先把睡眠时间保证好，再逐步把作息稳定下来，往前挪一挪。

平时碰到这种问题，我可能就搜搜 Google，或者直接问 AI，但查完之后，还得自己去判断这些信息靠不靠谱。

我觉得用 ScienceBuddy 比较方便的地方，就是能帮我解读这些英文论文，而且它生成的结果能够追溯到原文的具体位置。这��就更方便判断，AI 到底是在编胡话，还是确实有所依据。

这次核对时也发现，个别页码标得不准。不过它把要看的材料和对应的结论整理到一起了，确实省了不少来回翻找的工夫。

做生物医学研究，或者经常需要整理这类文献的朋友，可以试试。目前免费，浏览器打开就能用：
https://t.co/3P3SfwnM0c

#ScienceBuddy #PhAILabs #AIforScience

[Alfred Baudisch (Pardall) @AlfredBaudisch]

Pet project:
1) Advanced PS1/N64 game extraction (+all animations).
2) Website where you can browse and compare models, textures, animations, FMVs, for preservation.

Already have FF7, Mega Man Legends 1 and 2 and MediEvil fully ripped.

#FF7 #FinalFantasy #FFVII #PS1 #PSX

[Sadao Tokuyama @tokufxug]

高反射オブジェクトの3DGS「フローター（ノイズのように浮く粒）」を抑えるDAVINCI

🔹Killian Rando氏が開発
🔹メッシュ＋拡散・反射成分を分離学習
🔹異方性・球面ボロノイ＋NASGaborで外観を表現
🔹3DGSより高PSNR、容量は4分の1
🔹ラスタライズでリ���ルタイム描画

情報元はリプ欄。