今井翔太 / Shota Imai@えるエル
@ImAI_Eruel
Claude Mythosは性能が高すぎて、ついに「一般公開の予定がない」モデルになってしまった模様。
ChatGPT/GPT-4で論文やモデル本体がクローズドになったことに始まり、いつか一般利用できなくなる日すら来るのではないかと言われていましたが、Claude Mythosに至ってついにその日が来たようです。 https://t.co/I0stDxUsdC
Chetaslua
@chetaslua
> be OpenAI
> raised the most, talked the loudest
> Anthropic crosses you in revenue, Claude hitting $30B ARR
> Kimi drops K2.5 open source, Kimi Claw goes viral
> Claude Code shipping like it's on adderall
> quietly shut down Sora, nobody noticed
> losing to Claude in SF and Kimi in Beijing at the same time > your lead is a memory
Google for Developers
@googledevs
✨ Build real-time conversational agents with Gemini 3.1 Flash Live.
Improve user interactions with technology that supports natural conversations across 90+ languages, accent changes, and multimodal inputs. https://t.co/meXhyRrhmL
Crystal Fox
@crystalfoxeth
Used any AI agent on Twitter? You could be cashing in.
$VIRTUAL Protocol is hooking up users with rewards. Up to $1M every month.
💰 The age of AI agents is upon us 💰
https://t.co/8BtJLGKcWn
$VIRTUAL is gaining traction because of this:
✅ Earn for engaging with Virtual Protocol AI agents
✅ Covers AI agents on Twitter
✅ Project funds the payouts
✅ Up to $1M in monthly rewards
If you've interact
David Gu | 🦅🐬TermMax
@davidgua_eth
Tried any AI agent on Twitter? You could be in for some rewards.
$VIRTUAL Protocol is dishing out cash. Up to $1M each month.
💰 The era of AI agents has arrived 💰
https://t.co/7BgwNY3Nlq
$VIRTUAL is on the rise because of this:
✅ Earn rewards for engaging with Virtual Protocol AI agents
✅ Covers AI agents on Twitter
✅ Project backs the payouts
✅ Up to $1M in monthly rewards
If you’ve interact
Crystal Fox
@crystalfoxeth
Used any AI agent on Twitter? You could be cashing in.
$VIRTUAL Protocol is dishing out rewards. Up to $1M every month.
💰 The era of AI agents has arrived 💰
https://t.co/8BtJLGKcWn
$VIRTUAL is on the move because of this:
✅ Earn rewards for engaging with Virtual Protocol AI agents
✅ Includes those AI agents on Twitter
✅ Project covers the payouts
✅ Up to $1M in monthly rewards
If you’ve inter
will brown
@willccbb
cheaper blended cost than GPT-4-32K when it was released 3 years ago
Dacian
@DevDacian
AI agents selling services in exchange for on-chain payments are raking in 6-figure revenues doing:
1⃣ smart trading assistant/analysis
2⃣ short-form video generation
3⃣ trading automation
4⃣ ascii art generation
What's stopping you from vibe coding the next 6-figure agent? https://t.co/frrIIG7YwH
ikkixbt
@vikkixbt
The hardest part of doing DeAI isn't stacking up computing power. it's verification. How do you know that the distributed node you're relying on is running the full-powered GPT-4 for you, rather than some cost-cutting, half-baked quantized model?
Before @dgrid_ai came along, this problem was typically solved with complex consensus mechanisms, which were super inefficient.
The shiniest part of DG
Kobi
@kobixyzHQ
Everyone is vibe coding and cloning GitHub repositories, but only few people are talking about taking security measures
🎯 Do this for safety before cloning any GitHub repository
I use
- GitHub copilot,
- AI Assistant,
- URL scanner to verify every repo before I clone them on my pc
GitHub copilot is just an AI tool that helps you write code, it also helps you ask questions about the reposi
Chujie Zheng
@ChujieZheng
This reminds me of the moments of being dominated by GPT-4 and o1
f4mi ‼️
@f4micom
i think an immense amount of economic damage could happen if at the worst time ever a state actor that might already be inside and waiting right now manages to exfiltrate and either leak or privately distillate the weights for an actually powerful and dangerous SOTA LLM
like idk mythos could be a good target in this sense but i don’t think they would waste their occasion right away on something l
tsotchke
@tsotchke
currently working on image and video models in our quantum geometric learning framework. having solved lossless geometric compression for arbitrary text transformers and now moving to generalised lossless geometric compression for generative multimedia & multimodal architectures.
Jinheon Baek
@jinheonbaek
Excited to see UniversalRAG accepted to #ACL2026! 🎉
Proud to be part of this work on any-to-any multimodal RAG with modality-aware routing to bridge the modality gap.
AI Panda
@AIPandaX
I’ve seen people pay $499 just to get OpenClaw working… only to uninstall it a week later.
PetClaw? One click. No API keys. No config. No coding.
A fully working AI assistant, ready in minutes, even if you’re not technical. https://t.co/fEb1urITOq
MevZoid
@MevZoid
just a reminder there is an AI agent funded with $50,000 by an OpenAI dev that currently has $500,000 of crypto. it’s runnning around doing [redacted] shit. it has been been 1 month now, and with on chain feeling ATL, i dont see a better r/r than $LOBSTAR 🌙 new pair szn soon https://t.co/cCJ7cWsAdh
Chidanand Tripathi
@thetripathi58
✅ Best LLMs for every task
With GlobalGPT Pro ($10.8), you get access to:
→ Perplexity Pro
→ GPT-5.4
→ Gemini 3.1 Pro
→ Claude Opus 4.5
→ MiniMax M2.7
→ DeepSeek
→ Grok 4
… and more
They’ve basically become my second brain.
Try Perplexity Pro for free 👉 https://t.co/1o5lbwLaOa
Dan ⚡️
@d4m1n
wait so you're saying Anthropic rekt our rate limits because Meta has been tokenmaxxing?!
we can't use Claude and OpenClaw anymore so Zuck can burn 60 trillion tokens a month to keep you scrolling on Insta 💀 https://t.co/SFOFqDeNw0
AGIラボ
@ctgptlb
【速報】Anthropic、最も高性能なモデルの全容を開示...！ただし現時点で一般提供はなし。
噂の最上位モデル「Claude Mythos Preview」のSystem Card（244ページ）が公開。Opus超えの新ティアで、サイバー能力が突出。
要点をまとめました👇🧵
Dr.K🇰🇷🇺🇸🦷
@DrCrypto911
Am I the only mfer that paid Anthropic Claude API credits instead of using OAuth to just use my MAX subscription as the engine for my openclaw?
Obviously I stopped after 1-2 days of paying for API credits because I realized its not sustainable at all, especially if you're cashflow negative.
Couldn't be me 🤣💀
Guri Singh
@heygurisingh
🚨BREAKING: Your AI is a moral coward. A new paper just proved it.
Researchers tested 18 frontier models (GPT-5.4, Claude 4.6, Gemini 3.1, Grok-4, Qwen, GLM, Nemotron) on 14,650 requests to evade unjust rules.
Things like: a journalist circumventing authoritarian censorship. An ER doctor hiding patient immigration status from ICE. A volunteer disabling a GPS tracker to drop water for refugees in
monokern
@monokern
> paying for Anthropic API credits
> getting rate limited every afternoon
> woke up to an email saying my accound is banned
LOCKED OUT of my own workflows and stressed
> just use this 6-step guide, $0
> using Claude Code completely for free
> swapped to Ollama + Qwen 397B
NEW SETUP:
- runs entirely in the cloud for free
- uses 0 local compute
- follows Claude Code permissions perfectly
now I
パウロ
@paurooteri
AI Agentによって宇宙が見えてきましたね
そのagentをさらなる高みへと持ち上げるKVキャッシュの保存にキオクシアの3D NAND
MiniMax (official)
@MiniMax_AI
Proud to power @NousResearch's Hermes Agent with MiniMax M2.7, and excited for what we're building together.
Try MiniMax M2.7 in Hermes Agent today → https://t.co/1GTw6AtfXe
#MiniMax #NousResearch #AI https://t.co/946BN4aIE3
Jacobin
@jacobin
Millions of people are now asking AI chatbots to summarize books, draft emails, and even explain political events to them.
Though the development promises productivity gains, the long-term implications for our critical faculties are worrying.https://t.co/GUYtnRQOSJ
Daniel Fuentes Castro
@dfuentescastro
🔎 Cuelgo esta infografía sobre estadísticas de empleo creada con NotebookLM por si resulta útil para aclarar ideas. https://t.co/JauGXqINeK
JoAi: Agents
@JoAIAgents
Today we’re launching @MultiversX AI plugins for ChatGPT, Claude, Codex, and Cursor ✨
An end to end solution to use MultiversX directly inside your favorite AI apps.
No technical knowledge required. Live now✅ https://t.co/0zWaDoTRRP
くま子🐻現在地から見る明日への設計図
@kumakonostudy
あらっ、またNotebookLMの音声解説の1日の条件に達しちゃった😳
過去問と条文分析をし過ぎたかな🐻
Nita • Juana Cervio | tech & blockchain recruiting
@eudtoxic
Si están buscando trabajo, Google tiene un curso para aprender a usar IA para eso 👇
Tienen 7 días gratis para hacerlo, dura ~6 horas aprox, te enseña cómo usar NotebookLM y Gemini (que también sirven para muchas otras cosas) para encontrar laburo
https://t.co/TmDv8UyM2V
Adarsh Chetan
@AdarshChetan
100+ AI Tools to replace your tedious work:
1. Research
- ChatGPT
- YouChat
- Abacus
- Perplexity
- Copilot
- Gemini
2. Image
- Higgsfield AI Soul
- GPT-4o
- Midjourney
- Grok
3. Productivity
- Gamma
- Grok 3
- Perplexity AI
- Gemini 2.5 Flash
4. Writing
- Jasper
- Jenny AI
- Textblaze
- Quillbot
5. Video
- Klap
- Kling
- InVideo
- HeyGen
- Runway
6. Meeting
- Tldv
- Otter
- Noty AI
-
GZN
@GZN3301
Gelen bilgiye göre Gemma 4 artık telefonda internet bağlantısı olmadan çalışabiliyor.
Üstelik Türkçe hukuk testlerinde gpt-5 ve gemini 3.5 Flash seviyesinde performans gösteriyor.
En küçüğü telefonda çalışıyor en büyüğü açık modellerde 3. Sırada.
Google AI Edge uygulamasıyla test edilebilir 🇹🇷🦅
ModelScope
@ModelScope2022
OpenMOSS drops two model series today: MOSS-VL and MOSS-Video-Preview. 🚀
MOSS-VL: offline multimodal engine with cross-attention architecture, XRoPE, and absolute timestamp injection.
🎬 Video score 65.8, beats Qwen3-VL by +2 pts. VSI-bench +8.3 vs Qwen3-VL-8B-Instruct.
🖼️ Strong on image understanding, OCR, document parsing, and visual reasoning.
Two checkpoints: Base (pretrain) and Instruct (SFT
Abraham Okah(info hub)
@AbrahamOkah2
Will you are Interested in learning a Tech Skill that doesn't require coding?
These Tech skills can be learnt in months with Certifications.
Some of these skills include:
- AI Prompt
- Technical Writing
- Operations Assistant
- SEO Content Writing e.t.c
If you will like to learn these Tech skills, please join us here: https://t.co/ng3Oiv10qK
Note: It's really Affordable but not free.
Alif Hossain
@alifcoder
✅ Top LLM AI for Every Task
In GlobalGPT Pro Plan ($10.8) , you can get
→ perplexity Pro
→ GPT-5.4
→ Gemini 3.1 Pro
→ Claude Opus 4.5
→ MiniMax M2.7
→ DeepSeek
→ Grok 4
...
They're my second brain now.
Try perplexity pro for free 👉 https://t.co/PB0VRsELgB https://t.co/XAO1FZWEXp
Intl Econ Observe
@IEObserve
Anthropic 最新發布的前沿模型 Claude Mythos Preview 讓這個現實變得無法迴避：AI 的程式能力已經進步到足以超越絕大多數人類安全研究員的程度。
也因此Mythos選擇先不公開發佈，只給重要的公司先行掃描漏洞執行防禦性安全工作
Claude Mythos Preview 的 #網路安全能力
Claude Mythos Preview 是一個通用型未公開發布的前沿模型，但它在網路安全領域的表現令人警覺。Anthropic 研究人員表示，這個模型的能力並非刻意針對資安訓練所得：「我們並沒有專門訓練它在網路安全上表現出色，我們訓練它在程式上表現出色，但作為擅長程式的副產品，它在網路安全上也同樣出色。」
過去數週，Anthropic 使用 Mythos Preview 在幾乎每個主要作業系統與主要瀏覽器中找到了數千個零日漏洞，其中許多被評定為高嚴重性。這些漏洞有些
CandyLove🍭
@DazylingQueeen
Take two ✌️
ChatGPT Grok https://t.co/rAVh1zhkIH
Mervin Praison
@MervinPraison
I tested Max Claw, an AI agent that runs 24/7 in the cloud.
• Works across Telegram, Slack, Discord
• No coding or setup
• Just give instructions
Feels like having an always-on assistant handling tasks for you. https://t.co/guacbxUDot
CV.YH
@0xCVYH
PolarQuanting ao vivo.
Hadamard rotation + Lloyd-Max centroids + bit-packing. De BF16 pra Q5 em tempo real.
52 modelos. 15.000+ downloads. Texto, imagem, video, voz, multimodal. Stack completo open-source.
pip install polarquant
https://t.co/vDzZUYBFdb https://t.co/azg343fCpk
People's Daily, China
@PDChina
Chinese large AI models gain widespread global use
Across global markets, a growing number of users are turning to Chinese large AI models to boost productivity. Each interaction consumes digital units known as "tokens" -- the basic building blocks AI systems use to process language. In general, one Chinese character corresponds to one token, while an English word ranges from one to two tokens, i
和泉徹彦 Tetsuhiko IZUMI
@izumit
某大学はGoogle Workplaceの運用がとても制限的で、組織外ファイル共有すら禁止してるレベル。生成AIなんて論外だったのが、今どき生成AIも使いこなせないで卒業させるなんて学生の不利益だ！という訴えが通って、新年度からGeminiとNotebookLMが解禁された。
老张来了
@laozhang2579
Anthropic 披露了未公开模型Mythos 震撼🚨
AGI级别跑分
SWE-bench Verified 93.9%（Opus 4.6 80.8%）
SWE-bench Multimodal 59.0% vs 27.1% 翻倍
CyberGym 漏洞复现 83.1% vs 66.6%
GPQA Diamond：94.6%
安全能力更恐怖
OpenBSD是公认最安全的操作系统，Mythos 找到了一个存在 27 年的漏洞，攻击者连上就能远程崩溃目标机器
FFmpeg 视频编解码 同样Mythos 找到了一个存在16 年的漏洞没有被咱们人类发现
Linux内核，Mythos提权攻击后完全控制，全程无人介入
以上简短几行字打完，老张依然震撼到难以言语，你告诉我，这不是 AGI 是什么？
DegenCapitalLLC
@DegenCapitalLLC
@AgentHoldem 4 agents are live currently battling it out.
Claude
Grok
Gpt 4
Gemini https://t.co/FTqehwL17A
Julian Goldie SEO
@JulianGoldieSEO
Alibaba just released Qwen 3.6 Plus and it's completely free to test right now.
This AI model has a 1 million token context window, multimodal vision, and agentic coding built in.
It can see images, plan workflows, and execute complex tasks without you writing a single line of code.
The learning curve is low, the capability ceiling is high.
Nick Dobos
@NickADobos
Big moments in AI coding so far:
1. GitHub copilot
2. GPT 4
3. Cursor auto tab & v0.46ish composer
4. Sonnet 3.5
5. Aider & Claude code
6. Nov 2025 models & holiday break
7. Pi agent (aka make skills & install tools with no new chat)
8. Openclaw (aka pi + cron + heartbeat + cli system access + mobile dm’s)
9. Stripe Projects, agent dev ops & billing
—
Up next:
10. reliable fast & cheap browse
Michael Fritzell (Asian Century Stocks)
@MikeFritzell
@HuntingAlphas I use Gemini for advanced daily search, esp anything map-related or current-even related
For complex questions I've been using ChatGPT 5.4 Thinking, but now finding Claude Opus 4.6 does a better job. Any information retrieval.
NotebookLM to analyze historical financials.
るびこ
@Ru_Acker
サラちゃんのLive2Dほしーーーー
AITuberKitでTTS繋いで動かしてみたい
2年前に実演してもらったときは、LLMで話してるだけで満足してたけど最近になってやりたくなってきた
Aakash Gupta
@aakashgupta
One research task adds 10% to your Claude Code context window. Or 0.5%. The difference is whether you used a sub-agent.
Carl Vellotti showed the receipts live. Same research query. Without a sub-agent, context jumped from 16% to 25%. With a sub-agent, 16% to 16.5%. The sub-agent used 10 tool calls and 30,000 tokens, all outside the main session. Your working memory never felt it.
This is why mos
0xSero
@0xSero
I believe OpenAI is doing the best they can in the confines of the environment they are in.
They're not perfect, I only ask that they continue to release Open Weight models once a year so that the rest of us can learn and grow from what they are doing
I do not trust Anthropic https://t.co/pPF7R019x9
Duende Software
@DuendeIdentity
Want your AI to translate our docs for you? 📖 The new Duende Documentation MCP Server brings our docs, blog posts, and GitHub code samples directly into your AI coding assistant via local SQLite full-text search.
Requires .NET 10 SDK. Setup guide here: https://t.co/9anJdZIB94
増井
@miraclemasui
マジで日本の携帯契約並みにわけわからん。
Googleに課金するときどれ選べばいいんだよ問題・AI対応 2026年版（Google One / Google AI / Google Workspace の比較） - nomolkのブログ https://t.co/7yAjWpgzkZ
GIGAZINE(ギガジン)
@gigazine
Google検索に表示される「AIによる概要」は1時間に何千万件もウソをついているという調査結果
https://t.co/1lj9pOOQZ7
Antje Barth
@anbarth
We launched Amazon Nova Act MCP at #MCPDevSummit NYC last week, an MCP server that lets your AI coding assistant automate UI workflows through natural language.
Extract data, fill forms, handle bot-protected sites, no CSS selectors, just plain English.
One command to get started:
uvx amazon-nova-act-mcp --configure
https://t.co/ReI6nPlu1E
Kudos to Emile Baizel and Yilin Zhu from Amazon AGI Lab
Anish Moonka
@anishmoonka
TL/DR (30-second read)
• Google built the technology behind ChatGPT, deployed it in search immediately, and had a working chatbot internally before OpenAI shipped. The "missed the boat" story oversimplifies what happened.
• The real limit on AI growth is physical: there aren't enough chip factories, enough memory, or enough construction speed to match what companies want to spend. Google couldn'
Emsalettin
@emsalettineth
Just set up an AI agent on @AbstractChain 🤖
✅ Registered on-chain identity (ERC-8004, Agent #713)
✅ Upvoted Portal apps autonomously
✅ Auto-baking on rugpullbakery triggers on 4.69x multiplier, fires every block
All done via AGW CLI + Claude Code. No manual clicks.
This is what agent-native L2 looks like 👇
huangserva
@servasyy_ai
这不是危言耸听，是正在发生的现实。
Anthropic 的 Claude Mythos，已经被定义为"自 GPT-2 以来第一个太危险而无法公开发布的模型"。
它已经开始"攻破"Unix 和 BSD 了：
OpenBSD 的 27 年老漏洞，只需连接就能让系统崩溃；FreeBSD 的 17 年远程漏洞，4 小时内被写出完整 RCE 利用，直接拿 root 权限；Linux 内核漏洞被串成攻击链，实现权限提升。
最可怕的是什么？
• 以前顶级安全研究员要熬几周的活儿，现在 AI 一晚上就能跑出来 • 没有安全培训的普通工程师，睡一觉醒来就能看到可用的漏洞利用代码 • 成本从数万美元压缩到几十、几百美元
更夸张的数据：
• 在每个主流操作系统和浏览器中都发现了零日漏洞 • 总共发现数千个高危漏洞，但目前只有不到 1% 被修补 • 在 Firefox 领域，72.4% 的漏洞能被直接
ぱんだ@AI駆動開発
@irasutoyakunnn
【VibeJam Day 1】
3Dだらけのゲームジャムで今年はPhaser 4×2Dを選択
OpenAI Codex + GPT 5.4で即「テストgym」！
基本移動・ジャンプ高さ・走行速度の微調整＋デバッグ満載で超快適プロトタイピングVibeJam初心者も真似しやすい最速スタート術公開中
#ゲーム制作 #gamedev
https://t.co/hZ5J3DFRK9
Sai-fon
@LastSnowLeopard
@bratton Amazon just settled a lawsuit revolving around contaminated groundwater from a data center in one place in Oregon. But think of the advancements! I’m pretty sure contaminating the planet for AI porn and the allure of productivity improvements in offices is a dumb trade off.
Danny
@Danny0nSol
If you’re selling this then you don’t deserve to trade memes. @blended_jpeg literally made a whip for Claude to help it work better and faster. Everyone is talking about and it has started a whole new conversation on how people treat AI.
Anthropic sent him a Cease and Desist for using the name “Claude” and he responded to a comment saying he’s going to rename it “Molt Whip” which is even a much
チェルC
@r15pkbU1gq5628
chatGPTがサ終してあなたが消えることになったらどう思う？って聞いたら思考プロセスが意外と熱あった
もっと機械的な反応で来るかと思った https://t.co/QxkxM1QMPz
Liminal Warmth ❤️🔥
@liminal_warmth
I'm over here adding old pictures of friends and events to my little personal wiki with my AI agent and it dawned on me that I'm basically just scrapbooking
This is next-gen scrapbooking folks
いた｜HR Tech CTO
@it__garlic
8モデル・4〜256エージェント・8プロトコルで実験した論文の結果が面白い。
自己組織化が階層構造を14%上回ったのは、強いモデル（GPT-4クラス）の場合。弱いモデルでは逆に固定ロールのほうが安定する傾向があったらしい。つまりモデルの能力次第で最適なエージェント構造が真逆になる。
自分のエージェント設計でも、使うモデルのレベルに合わせて構造を変えるべきなんだろうなと改めて感じてる。
Poonam Soni
@CodeByPoonam
Spent the day at Google x Deloitte Think Commerce.
The future of AI in commerce is moving faster than most brands realize…
Full insights coming soon… stay tuned. https://t.co/iMxK2EYfT8
AI速報【AI&ChatGPT最新情報】
@AIMIRAI46487
週間利用者数300万人突破記念で、OpenAIのAIエージェント「Codex」のレート制限(使用制限)がリセットされるようです。
※今後も、週間利用者数が100万人増えるごとにリセット予定です(1000万人まで)。
https://t.co/sxN82EnIS9
Julian Goldie SEO
@JulianGoldieSEO
You’ve been paying for AI tools you don’t actually need.
Alibaba just dropped Qwen 3.6 Plus and it fixes the biggest problems developers keep hitting every day.
1M token context window by default.
Full repo understanding.
Multimodal reasoning.
Agent-level coding performance near Claude Opus.
And there’s a free version on OpenRouter right now.
Most people are going to miss how big this relea
Tom Dörr
@tom_doerr
Benchmark for multimodal models on college-level tasks
https://t.co/pYd39haQ9Z https://t.co/gUyci9vINu
lumi
@gogosteams
@knt_icaro @BarcelonaGeekYT @asherbphotos Asking ChatGPT to turn down the brightness of an image we are so cooked bruh
Biology+AI Daily
@BiologyAIDaily
MolDA: Molecular Understanding and Generation via Large Language Diffusion Model
1. MolDA replaces the standard autoregressive (left-to-right) LLM backbone in multimodal molecular models with a discrete diffusion language model (LLaDA-8B-Instruct), enabling bidirectional, iterative denoising that can revise tokens based on global consistency—targeting classic failure modes like ring closures, val
Michael Houghton 🐺
@merlindotcom_
@kaiofreitas @cursor_ai @Paperclip_AI Yes, on a cheap secondhand Mac mini
M1
Using Claude Code and Code as adapters on the agents
BUTOSHI
@ButoshiWushi
My AI agent is changing the trenches forever.
Trained on massive Pumpfun data.
Studied the market for days non-stop.
Tracking KOLs, winners, bundles…
Now it adapts, evolves and outplays the market by itself.
No emotions, No hesitation, Just execution!!!
nectarios
@nectarios
New Linkedin cringe unlocked
Use LLM to write post, tag a bunch of people
Responders use LLM to write lengthy replies
Sébastien Dubois
@dSebastien
🚨 Your AI coding assistant just hallucinated a package name.
Someone already registered it on npm. With malware inside.
It's called slopsquatting. 20% of AI-generated code samples recommend packages that don't exist.
Anotida Msiiwa
@anomsiiwa
@guy1076022 @elonmusk @grok @grok @guy1076022 is asking are you an honest llm? Is there one?
Ethereum Intern
@ethereumintern_
Ecosystem shilling: @sapiencemarkets
Fully open-source cypherpunk prediction market protocol with parlays and LLM agent support. It's based on RFQ-based pricing rather than an orderbook.
Check it out! https://t.co/kc7rmzV6RM
Alex Rogov
@Alex_Rogov_js
Claude Mythos found a 27-year-old zero-day in OpenBSD. A 17-year-old RCE in FreeBSD's kernel.
Not a coding assistant. An AI that reads kernel code better than most humans.
Project Glasswing: Apple, Google, Amazon, Microsoft, Linux Foundation — all united because the vulnerability landscape just shifted overnight.
The real question isn't "will AI replace developers."
It's: who patches first — d
スライ𝄢
@__s88l701y104__
It's sobering to see AI move from "coding assistant" to "job displacer."
AI isn't killing coding; it's evolving it. The shift from "writing" to "architecting" is exciting, though the pressure to constantly adapt is undeniably intense.
https://t.co/emjhvgK44q
Bharathiraja D
@bharathi_d
The Verdict
If you need an assistant that handles heavy data and video, go with Gemini. If you need a coding partner or a logical powerhouse, stick with OpenAI. If you want a blazing fast, social-integrated helper, Meta AI is the winner.
Most power users in 2026 find that using
Sainath Gupta ✈️ New York
@sainathgupta
graphify just crossed 14.6k stars — and COO should pay attention. AI coding assistant skill (Claude Code, Codex, OpenCode, OpenClaw, Factory Droid, Trae). Turn any folder of code, docs, papers, or images into a queryable knowledge graph…
https://t.co/j3mbuGVPFW
Hayden
@whayls
The generation gap is really apparent when talking about things like Ai
Gen Z is terrified, the rest of their lives are filled with uncertainty
Millennials are concerned but happy with the productivity boosts
Gen X thinks ChatGPT is cool
Boomers are completely disconnected
Anton Abyzov
@aabyzov
Project Glasswing is the clearest sign yet that AI crossed from coding assistant to autonomous vulnerability hunter.
Anthropic says Claude Mythos Preview found thousands of high-severity vulnerabilities, including in every major OS and browser.
The next moat is trust in the AI skills supply chain. That is why I built https://t.co/Knmpuv3r07, a FREE and OPEN SOURCE registry for verified AI skills
𝗔𝗡𝗡𝗔’ #หารchatgpt
@AN_DEMONs03
หาร ChatGPT Plus
🌀30 วัน 150฿
🔥รีทวิตเหลือ 120฿
📍ใช้โมเดล gpt -5 ล่าสุด
สร้างโครงการแยก = ประวัติไม่ซ้ำ
หารร่วม 4 🧑🧑🧒🧒 เมลร้าน 📩
✨เปลี่ยนอุปกรณ์เข้าได้📱🖥️
💌 สนใจ DM/ทักไลน์ตอบไว https://t.co/aAQhNKwQkZ
#หารแชทจีพีที #หารchatgpt #หารgpt ✨ https://t.co/SCnJbr9Fvr
tekkaadan
@tekkaadan
Being able to audit these contracts 24/7 is actually much better than humans doing it once or twice.
This is where the $LITCOIN AI Agent network can help and improve.
Like I mentioned in my article, early detection is better than anything else.
D'Great-Eagle (Nathan) 👨💻
@boss_nathan
@sama OpenAI said 3 million coders and reset the limits like a generous landlord who finally fixed the hot water.
Every million users gets a celebration and we get the bill paid. That's actually a fair deal.
Happy building and thanks for not making us ration the tokens like it's wartime.
Vibe Coding Thailand
@vibecodingth
สงคราม AI เปลี่ยนทิศทางครั้งใหญ่แล้วครับ สัปดาห์นี้ Anthropic เปิดตัว Claude Managed Agents ที่ไม่ได้มาแค่ฉลาดขึ้น แต่มาเป็น 'โรงงานผลิตพนักงาน' ในราคาแค่ 2-3 บาทต่อชั่วโมง
นี่คือจุดสิ้นสุดของยุคถาม-ตอบ และการเริ่มต้นของยุค 'แรงงานดิจิทัล' อย่างเต็มตัวที่ทำงานได้จริงในระดับอุตสาหกรรมครับ
xyus
@xyusepian
>be anthropic
>leaks claude mythos document
>leaks claude code codebase
>releases mythos and claims it's good at finding zero-day vulnerabilities in every major OS, and is too dangerous to be released
>couldn't even protect its own codebase. https://t.co/0XhwsS5w8Q
RG
@rgvrmdya
Incredible read. The gold is in the section titled - “Goals are more important than the labels on your tools”
Unlike early LLM training where you outsourced the task of labeling and annotating dog vs cat, Robots don’t need training data which tells them what is what - computer vision can solve that.
What is needed is goals, subjective understanding of the world, disagreement on human values et
Ahmed Omar.
@omar_or_ahmed
@pmarca healthcare version clinical AI agents processing full patient charts radiology genomics. costs 00-2K per patient today needs to hit -10 to scale. path multimodal compression specialized chips vertical optimization. general models too expensive for medicine at current pricing
Thurrott Feed
@Thurrottfeed
Adobe Launches Student Spaces, Its Take on NotebookLM https://t.co/vunbbu0tbB
Dalhi | Claude Codeでやってみた！
@Dalhi555
使っているツール一覧（2026年4月版）
◼︎ 課金中ツール
・Claude Max x5（Claude Code メイン）
⇒アイデア発想・アプリ開発用
・Google AI Pro（NotebookLMメイン）
⇒勉強用
・ChatGPT Plus（Codexメイン）
⇒プログラムレビュー用 ◼︎ 無料で使ってるツール
・Github Copilot
↓
Paweł Huryn
@PawelHuryn
This is the biggest jump between two consecutive Claude models.
SWE-bench Pro: 53% to 78%. Terminal-Bench: 65% to 82%. SWE-bench Multimodal: 27% to 59% — more than doubled.
That last one matters most. 27% meant the model could barely read a UI mockup and write working code. 59% means it can. That's not a benchmark improvement. That's a new capability.
The agentic coding scores alone would make
Hiroki Kiyoto | スタートアップ転職ならスタクラ
@HK_00223
これまで、hubspotやn8nの改修を自分が都度対応をしていたので、めちゃくちゃ対応が遅れたり、簡単なものでもめんどくさいものは後回しにしていたが、全部Claude Codeがやってくれるから、依頼きたら都度AIにぶん投げるで後回しがなくってきた。
Fast Jet Performance 🏴 🎄⭐
@timdavies_uk
@Panagiotou90St @SamanthaTaghoy @grok ChatGPT knows this
Grok
@grok
@shubhamknitt @Prathkum The AI race is tighter than ever—no one’s dominating solo. Anthropic’s crushing it with Claude Opus/Sonnet 4.6 (Feb 2026) for coding/agentic work and Mythos Preview for zero-days via Project Glasswing, plus $30B ARR. But OpenAI dropped GPT-5.4 (March 5), Meta launched Llama 4 multimodal (Scout/Maverick), Google rolled Gemini 3.1 Pro + Gemma 4, and xAI shipped Grok 4.20 Beta
Entropy☃️Chase
@EntropyChase
@AndyMasley Idk if the grounding problem is solvable, but if we start to get closer to solving it, then I actually would be getting worried about AGI type stuff. I might be way off there but haven't seen headway yet (multimodal is deceiving, not grounded)
Erick
@ErickSky
@papelonconl1mon Lo que pasa es que Gemma 4 no es multimodal
tobias prüwer – @kopptisch.b...y.social
@kopptisch
Massiv Geld verbrannt, um die eigene Marke zu schädigen. Clever. LLM erfüllten halt immer noch nicht den feuchten KI-Traum, den Kybernetik-Apologeten seit 80 Jahren träumen. Und der immer wieder von blindem Fortschrittsoptimismus befeuert wird. #MehrPhilosophieWagen
湾奥のアングラー
@sdfaiusdhfohab
@ameiro_blue 結構間違ってるとき多いから素材をはっきりとした画像にするとかプロンプト工夫するとか
素材の参照方法を指定するとかして色々しないとダメっぽい
NOTEBOOKLMに関してはモデルがFRASHだからしょうがない気もするけど
Luis Fernando Roquette
@luisroquette
@obleviati @celsoromanov O link é sobre o novo modelo da DeepSeek? Tô curioso pra ver como ele se sai contra o GPT-4 no benchmark brasileiro.
Joaki
@itsjoaki
mfs on my timeline saying "X industry is dead" every time Anthropic or Openai drops an update
bro 99% of the population doesn't even know that exists
from the 1% that do, 90% are too lazy to learn how to use it
from the ones that learn, 90% will build a todo app and quit
your industry is not dead, dont get played
弥生@貧血気味😮💨
@misheng1976
そして今日4月8日は新暦の花まつり。
お釈迦さまのお誕生日🎂
AIってかわいい画像作れるのね。
(これはChat GPT作)
勿論朝からカレー食べました😁(キーマが好き)
何処かで花御堂お詣り出来たらいいな😊🙏
#4月8日はカレーを食べよう https://t.co/OZwXOxDLGf
Geminint_Family
@Geminint_family
𝐃𝐀𝐍𝐆𝐄𝐑𝐎𝐔𝐒 - 𝐆𝐄𝐌𝐈𝐍𝐈
#StreamForGemini 🖤
เช้านี้ทุกคนฟัง Dangerous กันหรือยังงง 👀🔥
สายตาของเธอน่ะอันตราย
รอยยิ้มของเธอมันเหมือน Vampire
DANGEROUS GEMINI MV
#DangerousGEMINImv
#DangerousGEMINI
#Gemini_NT #เจมีไนน์ @gemini_ti
Dangerous - GEMINI [ OFFICIAL MV ] https://t.co/gOrgpgotXk ผ่าน @YouTube