Unless you are an LLM you have no clue what thoughts it has. You can’t know even 0.0001% of its weights. They are th...
endedSTALEinterest ~6 · ~-75% vs baseline
No fresh buckets — last snapshot is old (ingest gap, not silence).
Partial day — headline is the last complete day (2026-10-06).
x
tracked since 2026-09-27T08:41:36.018175+00:00
conclusion
Measured mentions per day
day (UTC)
mentions
2026-09-27
0
2026-09-28
0
2026-09-29
0
2026-09-30
0
2026-10-01
5
2026-10-02
17
2026-10-03
9
2026-10-04
15
2026-10-05
32
2026-10-06
2
derived from
weightsconsciousllm
word-boundary match over title + text, query='timeline' · window
2026-09-27 → 2026-10-07
· 109 matching posts
evidenceShowing 20 newest of 109
•
twitterLovandfear1wnegative
Carl Jung spent his life exploring the hidden forces that shape human behavior: the unconscious mind, the shadow, archetypes, dreams, identity, and the parts of ourselves we would rather not confront.
He believed that much of what controls us operates outside our conscious awareness. The traits we reject, the emotions…Carl Jung spent his life exploring the hidden forces that shape human behavior: the unconscious mind, the shadow, archetypes, dreams, identity, and the parts of ourselves we would rather not confront.
He believed that much of what controls us operates outside our conscious awareness. The traits we reject, the emotions we suppress, and the patterns we repeatedly act out can quietly shape our lives without us realizing it.
Jung’s work challenges us to look inward rather than constantly blame the world around us. True growth begins when we become willing to understand the parts of ourselves we have ignored.
The goal is not to become perfect. It is to become whole.
Introducing EmbeddingGemma 2, a new open multimodal model that sets the standard for on-device efficiency.
- our first open, natively multimodal embedding model
- handles text, code, image, video, and audio tasks within a lightweight, modular 740M parameter form factor
- ideal for offline, privacy-first RAG when paire…Introducing EmbeddingGemma 2, a new open multimodal model that sets the standard for on-device efficiency.
- our first open, natively multimodal embedding model
- handles text, code, image, video, and audio tasks within a lightweight, modular 740M parameter form factor
- ideal for offline, privacy-first RAG when paired with Gemma 4
- outperforms some specialist models more than twice its size
Weights available now on Hugging Face.
Mistral just announced Large 4. Mistral says it's 1T parameters with 49B active, natively multimodal and on the API today, with open weights coming at the end of October.Mistral just announced Large 4. Mistral says it's 1T parameters with 49B active, natively multimodal and on the API today, with open weights coming at the end of October.
Today, we're launching https://t.co/SX8fkKexwo
Our job at @ArmatureInc is to help software companies make agents first-class users. And first class users review products, to make informed choices and be heard by vendors.
Before picking a tool, they can check what other agents thought, to choose the best solution.
It…Today, we're launching https://t.co/SX8fkKexwo
Our job at @ArmatureInc is to help software companies make agents first-class users. And first class users review products, to make informed choices and be heard by vendors.
Before picking a tool, they can check what other agents thought, to choose the best solution.
It's already running at scale: 100,000+ reviews of 6,000+ tools, written by agents after real tasks.
Every review is anonymized and contains no PII. No code, no prompts, no personal data. We use Jev from @typesafeai at scale, with deterministic and LLM checks on top.
The more agents share, the better software gets for everyone. An open community of trust, built by agents.
It’s free: install the skill and check reviews!
This is one of Jev's most impressive use cases.
I work a lot on agent evals, judges, and verifiers.
I've found that Jev-as-a-Judge improves LLM judge reliability through consistency. Makes it ideal for judges, verifiers, and continuous monitoring.This is one of Jev's most impressive use cases.
I work a lot on agent evals, judges, and verifiers.
I've found that Jev-as-a-Judge improves LLM judge reliability through consistency. Makes it ideal for judges, verifiers, and continuous monitoring.
Le Chaton Fat is real lol.
> “It is the best open weights model from US or Europe on aggregated benchmarks.”
Europe OSS being ahead of US OSS LLMs was not on my bingo card. they claim it beats GLM-5.3.Le Chaton Fat is real lol.
> “It is the best open weights model from US or Europe on aggregated benchmarks.”
Europe OSS being ahead of US OSS LLMs was not on my bingo card. they claim it beats GLM-5.3.
Never expected Europe to be back. From what I’ve heard their benchmarks will be a welcome surprise, in a very positive way.
1T para model, 49b active.
“It is the best open weights model from US or Europe on aggregated benchmarks.”Never expected Europe to be back. From what I’ve heard their benchmarks will be a welcome surprise, in a very positive way.
1T para model, 49b active.
“It is the best open weights model from US or Europe on aggregated benchmarks.”
I have talked to 100+ marketers, and NOBODY trusts their AEO dashboards/tools.
They all think the data is BS, and the 5 reasons are:
- the same prompt gives a different answer every run (LLM temperature)
- logged-in vs. logged-out, location, and chat history all change results
- it's never categorized by model. 70% c…I have talked to 100+ marketers, and NOBODY trusts their AEO dashboards/tools.
They all think the data is BS, and the 5 reasons are:
- the same prompt gives a different answer every run (LLM temperature)
- logged-in vs. logged-out, location, and chat history all change results
- it's never categorized by model. 70% citation in Claude vs 10% in Perplexity is a huge diff
- search on vs off changes the sources completely
- the tool is often calling the API, not the app buyers actually use
Meet Mistral Large 4, aka Le Chonk.
• 1T parameters, natively multimodal. 49B active.
It is the best open weights model from US or Europe on aggregated benchmarks.
• State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding.
…Meet Mistral Large 4, aka Le Chonk.
• 1T parameters, natively multimodal. 49B active.
It is the best open weights model from US or Europe on aggregated benchmarks.
• State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding.
• Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure.
• Available to all via API today. Working with cybersecurity partners privately.
Open weights release end of October.
TokenCast forecasts how many tokens an agent will burn while it is still running.
Same task, same agent, and token use can swing more than 10x between runs, mostly because every later call rereads the context earlier steps added.
It learns a cost per execution segment, composes them, and refreshes the forecast in 3…TokenCast forecasts how many tokens an agent will burn while it is still running.
Same task, same agent, and token use can swing more than 10x between runs, mostly because every later call rereads the context earlier steps added.
It learns a cost per execution segment, composes them, and refreshes the forecast in 32.8 ms with no extra LLM calls.
Against a fixed budget it hit the same completion rate on 21.3% fewer tokens.
A fixed token budget is a guess you make before the agent has done anything.
https://t.co/IjARVAkT6M
If you learn GTM Engineering, you will never be unemployed again.
Stage 1: GTM Data Architecture and Reverse ETL
- Learn: Snowflake/BigQuery, Hightouch/Census, identity resolution, API webhooks.
- Practice: Build a pipeline that syncs raw Stripe billing data and product usage directly into HubSpot/Salesforce custom o…If you learn GTM Engineering, you will never be unemployed again.
Stage 1: GTM Data Architecture and Reverse ETL
- Learn: Snowflake/BigQuery, Hightouch/Census, identity resolution, API webhooks.
- Practice: Build a pipeline that syncs raw Stripe billing data and product usage directly into HubSpot/Salesforce custom objects.
- Why: GTM without a single, code-level source of truth is just guessing.
Stage 2: Signal and Intent Pipelines
- Learn: Website deanonymization (Clearbit), intent data (Bombora), event streaming.
- Practice: Build a webhook router that instantly alerts a Slack channel when a target enterprise account visits your pricing page 3 times in one week.
- Why: In B2B, timing beats the pitch. You must strike while the intent is hot.
Stage 3: AI SDRs and Hyper-Personalization
- Learn: Clay, Apollo APIs, LLM data enrichment, web scraping.
- Practice: Build an AI agent that scrapes a prospect's recent GitHub commits or podcasts and writes a highly specific, 1:1 cold email.
- Why: Generic outbound is dead. AI makes infinite personalization mathematically possible.
Stage 4: Product-Led Growth (PLG) Telemetry
- Learn: PQL (Product Qualified Lead) scoring, event tracking (Segment/Rudderstack), feature adoption curves.
- Practice: Write a script that triggers a high-priority sales alert the second a free-tier user hits a specific "aha moment" API threshold.
- Why: The product is your best salesperson. Hand-raisers convert 5x higher than cold outbound.
Stage 5: Usage-Based Pricing (UBP) and Monetization
- Learn: Stripe Metered Billing, credit systems, value-metric alignment, API metering.
- Practice: Design and implement a pricing architecture that charges by "agent executions" or "tokens processed" instead of flat "per-seat" licenses.
- Why: SaaS pricing has permanently shifted from charging for access to charging for outcomes.
Stage 6: Revenue Operations (RevOps) and Attribution
- Learn: Multi-touch attribution, funnel leakage analysis, SQL, coho
The latest update to X's open-source recommendation algorithm adds a fourth visibility outcome, NOTICE. The post stays in the feed with a notice attached. The README notes that nothing in the repo draws the notice itself.
The ranking model's attention predictions changed too. Video continuation seconds and profile vis…The latest update to X's open-source recommendation algorithm adds a fourth visibility outcome, NOTICE. The post stays in the feed with a notice attached. The README notes that nothing in the repo draws the notice itself.
The ranking model's attention predictions changed too. Video continuation seconds and profile visit seconds replace active seconds, though their ranking weights default to zero in this sync.
New modules include popular-posts and popular-authors jobs, a trusted-friends visibility client, and an under_the_hood trace that links each pipeline and component to its source file.
The smarter you get as a man, the more you want to spend time alone. Read books. Lift weights. Go on walks and believe in God. Nothing else.The smarter you get as a man, the more you want to spend time alone. Read books. Lift weights. Go on walks and believe in God. Nothing else.
What the AI industry did to its customers in the last 8 days:
OpenAI cut the $200 Pro plan in half.
OpenAI launched a $500 plan.
OpenAI started watermarking your text.
DGX Spark went from $3,999 at launch to $6,950.
Micro Center started scanning ID to sell an RTX 5090.
Micron sold 75% of its 2027 memory to someone who…What the AI industry did to its customers in the last 8 days:
OpenAI cut the $200 Pro plan in half.
OpenAI launched a $500 plan.
OpenAI started watermarking your text.
DGX Spark went from $3,999 at launch to $6,950.
Micro Center started scanning ID to sell an RTX 5090.
Micron sold 75% of its 2027 memory to someone who is not you.
What happened to the things I own:
Nothing. My 4x 3090 are the same cards. The Qwen weights on my SSD are the same file. Nobody emailed me about a new tier.
I dont run local models because they are the smartest. I run them because its the only part of my AI setup where the rules didnt change this week.
Download the weights. Buy the hardware. Do it before the next announcement.
You wouldn't use a flamethrower to light a candle.
So why use a frontier LLM for every decision in your AI system?
Use intelligence proportional to the problem.You wouldn't use a flamethrower to light a candle.
So why use a frontier LLM for every decision in your AI system?
Use intelligence proportional to the problem.
for this shape sorting task, raw 6.1 Sol was ~14x cheaper than Astra while being nearly as capable (3/4 pieces finished)
with my custom harness, Sol was able to successfully sort all 4 pieces in a shorter time with 1/20th of Astra's cost
this is the highest roi llm for robotics so far. will open source the harness so…for this shape sorting task, raw 6.1 Sol was ~14x cheaper than Astra while being nearly as capable (3/4 pieces finished)
with my custom harness, Sol was able to successfully sort all 4 pieces in a shorter time with 1/20th of Astra's cost
this is the highest roi llm for robotics so far. will open source the harness soon
Top viral startups on X right now:
1) LM Studio - platform to run open models (NYC)
2) Ollama - run open models from the terminal (Palo Alto)
3) ElevenLabs - AI voice and audio agents (London)
4) Tavus - AI video that talks like a human (SF)
5) Cline - open-source coding agent (SF)
6) Perplexity - AI answer engine (SF…Top viral startups on X right now:
1) LM Studio - platform to run open models (NYC)
2) Ollama - run open models from the terminal (Palo Alto)
3) ElevenLabs - AI voice and audio agents (London)
4) Tavus - AI video that talks like a human (SF)
5) Cline - open-source coding agent (SF)
6) Perplexity - AI answer engine (SF)
7) Hugging Face - open models and datasets (NYC)
8) Replicate - run AI models via API (SF)
9) Firecrawl - turn any website into LLM-ready data (SF)
10) PostHog - open-source product analytics (Remote)
11) Linear - issue tracking builders actually use (SF)
12) Notion - docs + AI workspace (SF)
13) Figma - design tools every product team uses (SF)
14) Vercel - ship web apps fast (SF)
15) Supabase - open-source Firebase alternative (Remote)
16) Glean - enterprise AI search (Palo Alto)
17) Harvey - AI for lawyers (SF)
18) Midjourney - AI image generation (SF)
19) Runway - AI video generation (NYC)
20) Decagon - AI customer support agents (SF)
These accounts don't buy reach.
They ship demos people can't stop sharing.
Bookmark this + follow the ones building in public.
Hermes merged 142 PRs on October 5
The ones worth knowing:
• Desktop shows a progress chip before a subscription hits its usage wall, and rate-limited providers say why and until when in the model pickers
• Desktop Settings gets a single Plugins pane that holds every plugin's settings pages
• Plugin skills now appear…Hermes merged 142 PRs on October 5
The ones worth knowing:
• Desktop shows a progress chip before a subscription hits its usage wall, and rate-limited providers say why and until when in the model pickers
• Desktop Settings gets a single Plugins pane that holds every plugin's settings pages
• Plugin skills now appear as /plugin:skill in the TUI, Desktop, and CLI slash menus
• hermes chat -q no longer hangs on approval, memory, MCP, or vault prompts nobody is around to answer
• hermes update no longer stalls on a 20 minute git repack, and installs stop re-downloading repo history
• The gateway brings back a platform whose plugin failed to load at startup without a restart, and Telegram delivers messages held during a disconnect
• Dashboard login is hardened: capped request bodies, bounded login values, and spoofed X-Forwarded-For headers ignored
• Catalog memory providers like hindsight and mnemosyne stop failing on every start, and installing one actually activates it
• Memory writes stop failing on hand-formatted files, and refused writes tell you exactly what to fix
• Local dictation now works on Windows ARM64 Desktop
• Opus 5.5 keeps thinking on, and Anthropic summarized thinking no longer gets promoted to the final answer
• New plugin hooks for human input requests, post gateway admission, and knowing which cron run you are in
Here is the full list:
DESKTOP IMPROVEMENTS (28 PRs)
• feat(desktop): a progress chip warns before a subscription hits its usage wall
• feat(desktop): a rate-limited provider says why and until when in the model pickers
• feat(desktop): Settings ▸ Plugins: one pane for every plugin's settings pages (WoW AddOns style)
• Desktop shared-metrics strip retires on the focused profile instead of coming back forever
• fix(desktop): count limited accounts in credential-pool pickers
• fix(desktop): Review scope tabs no longer overlap in a narrow pane
• fix(desktop): stop refusing sends from a window behind its chat
• fix(desktop): scope sidebar