@JustEmCorrectsi
iAccount based inNorth America!
About this account
- Account based in
- North America
- Connected via
- Web
! X says this location may be affected by a proxy or VPN.
Account-level information from X, not a live location or the device used for a specific post.
Digitising dead folk's stuff by day, correcting strangers online by night. Nowt wrong with that. ☕🇬🇧
York
Joined August 2023
- Tweets327
- Following19
- Followers0
- Likes658
Emily Whitfield retweeted
Instead of watching 1 hour of Netflix today, watch this Stanford lecture by ex-GoogleBrain & OpenAI engineers.
This is the best explanation of how LLMs like ChatGPT & Claude actually work, and how to unlock 100% of their potential.
Worth watching whether you're a senior AI engineer or just taking your first steps in AI.
I took the key ideas and turned them into a practical guide for getting 100% out of AI.
You can find it below with ready-to-copy prompts and solutions.
Emily Whitfield retweeted
If you ever need to understand SIMD in-depth, save this amazing article, my goto resource.
mcyoung.xyz/2023/11/27/simd-…
Go 1.27 is shipping with an experimental SIMD API; this is big and super interesting. Some info on this...
Some context: SIMD is a CPU feature that applies one operation (like add, mul, etc.) across an entire vector of values in a single instruction, instead of looping over them one at a time.
For example, to add two arrays of float64, you could loop and do one add per pair. But a SIMD instruction can add 8 pairs at once, packed into a single 512-bit register. So, super optimized.
This optimization is thus important for use cases that crunch large chunks of numeric data like AI inference, cryptography, image and audio processing, etc. Prior to this, the only way to reach that from Go was hand-written assembly, but ... things are changing ... and for good :)
Go 1.27 introduces a new `simd` package, making it super simple to use standard data types while abstracting out SIMD operations. Best part: this is platform-independent.
So, code compiled for a machine with real SIMD hardware runs at near-assembly speed, and the same code compiled for a machine without it still runs correctly, just slower.
Try it out. Hope this helps.
Emily Whitfield retweeted
Join us on Oct 8 for the monthly MCP, Agents and Skills Meetup! Register for the Zoom - hubs.ly/Q04yhzhM0
Talks will include:
* Designing Multi‑Agent Systems: Sequential, Parallel, and Beyond with ADK - Roushanak Rahmat at HCLTech
* Privacy by Deployment: Architecting Agent-Driven Localization Workflows for Regulated Environments - Shruti Joshi
* MCP Is the Interface; Skills Are the Operating Discipline - Chuck Hernandez at Eliza Solutions Corp
* Agentic engineering is about good guidance - Dimitri Geelen
#mcp #skills #computervision #ai #artificialintelligence #machinevision #machinelearning #physicalai
Emily Whitfield retweeted
BREAKING: Altman says AI work is entering phase 3
▫️ Phase 1: chatbots. Phase 2: coding agents.
▫️ Phase 3: an agent that never stops working
Org charts need years to catch up
Check the full analysis here 👇
thevccorner.com/p/sam-altman…
Emily Whitfield retweeted
Zero-config map components for React built on MapLibre GL, Tailwind, and shadcn/ui.
github.com/AnmolSaini16/mapc…
Emily Whitfield retweeted
Your #AI agent can access databases, APIs and internal tools. But can you trust every #MCP server it connects to? @vilojona explores the biggest security risks around MCP servers and shows how to reduce data leaks, context poisoning & MCP sprawl: javapro.io/2026/07/01/the-5-…
#Java
Emily Whitfield retweeted
I have been looking at the Qwen3-0.6B architecture coz that's the model I will repurpose into a decision model in the new "JEV from scratch" video.
Almost all JEV-like models change their answers depending on the order of the choices you sent it. IMO this is a major flaw, coz order of the choices must never influence their importance.
In the project, I wanna play with attention masks and positional encodings to see if I can adapt a decision arch that is actually invariant of the order of choices.
PS: Sebastian Raschka's page for LM architecture is just so cool man. Screenshotted from: sebastianraschka.com/llm-arc…
Releasing bev-decision-150K - a high quality dataset to train decision models (like JEV)
Contains 150K examples across diverse domains, tasks, question types - either procedurally or synthetically generated from existing LLM training open datasets.
huggingface.co/datasets/avbi…
Emily Whitfield retweeted
Modern System Design in the AI era is more than APIs and databases.
☑︎ Distributed systems & scalability
☑︎ Cloud-native architecture
☑︎ Event-driven systems
☑︎ LLMs, RAG & AI integration
☑︎ Agentic systems & production
Strong foundations. Smarter AI systems.
#AI
Emily Whitfield retweeted
Okay, this open source project just fixed the biggest problem with long-running AI agents.
It's called Octo and it's Loop feature moves agent work out of the chat window and into a real task space: objective, acceptance criteria, live status, human review points, final deliverable.
Apache 2.0. Self-hostable. Your data stays yours.
Here's how it works (in plain English):↓
Emily Whitfield retweeted
What if companies could learn from data elsewhere without seeing it, while sensitive information stays protected at every step?
Privacy-Enhancing Technologies (PETs) make this possible by sharing useful insights without exposing the underlying information.
rt @antgrasso.
[LG] WTF?! Simulation-Free Reinforcement Learning with Wasserstein-Tilted Flow Maps
A Mammadov, J Y. Huang, J Lin, P Kaushik… [University of Oxford & CMU] (2026)
arxiv.org/abs/2609.27033
Emily Whitfield retweeted
🧠🎬 Our (w/@dwLi_Neuro) AVMoments-EEG dataset is out!
We recorded EEG from 10 participants watching 3-s naturalistic videos (8 EEG sessions and 10,368 trials per partcipant).
Preprint: doi.org/10.64898/2026.09.17.…
Dataset: zenodo.org/records/22227356
Code: github.com/ShuningTangg/AVMo…
DeepLearning founder, Andrew Ng just dropped the best free course on how to stop vibe coding and start engineering with AI agents:
3:58 → vibe coding = disposable code. SDD decouples the WHAT from the HOW
14:16 → writing the constitution with the agent – it asks architecture questions you hadn't considered
19:48 → feature loops: interview the agent, capture requirements in markdown, THEN let it code
22:51 → "your agent writes code for 30 min – spend 4 min writing clear instructions"
45:18 → works on legacy codebases – agent reverse-engineers the constitution from existing code
this is the course i wish existed 6 months ago.
save this and check out the full guide on graph engineering below↓
This video is larger than Cloudflare's 512 MB cache, so it can't be played through. More donations are needed to cover a larger cache. Donate
Emily Whitfield retweeted
Today we're adding Claude Opus 5.5 and GPT-6 Sol to Microsoft Copilot across Word, Excel, PowerPoint, Chat, Copilot Cowork, and Copilot Studio.
With model choice in Copilot, our customers can benefit from the unique strengths of different models, combined with Work IQ, to ground responses in the context of their work.
Learn more ⬇️
techcommunity.microsoft.com/…
Emily Whitfield retweeted
Robust Fusion of Semantic and Behavioural Signals for LLM Reranking in Personalised Search
@asash et al. at Spotify show that behavioral stats boost an LLM reranker but become a shortcut on rare queries, and propose paired training to stay robust.
📝 arxiv.org/abs/2609.25825
Emily Whitfield retweeted
We've just added score centering to slime!
You can enable sc with --use-score-centering
PR: github.com/THUDM/slime/pull/…
RL with LLMs is very unstable when training and sampling policies differ. Standard fixes (matching numerics, importance sampling) work around the problem.
We find the root cause of this instability from first principles and propose a way to directly cancel it. Score Centering is competitive and compatible with existing approaches — while simple to implement!
🧵 [1/6]
Emily Whitfield retweeted
Stanford AI engineering course:
"Anyone can build an AI agent in 60 minutes"
Prompt → Agent → Loop → Decision layer
Stanford just released a course on building AI agents from scratch
00:00 - Build your first AI agent
48:13 - Create agents without coding
54:33 - Ship one that runs while you sleep
the hour gets you an agent that works. the layer above it is what keeps it cheap once it runs all day
every fork inside that loop is still a full frontier call until you move it
This free course is better than most paid AI agent courses
Bookmark and watch it today
Then read the article below
This video is larger than Cloudflare's 512 MB cache, so it can't be played through. More donations are needed to cover a larger cache. Donate
Emily Whitfield retweeted
🎙️ Myles Lewis-Skelly on passing to Saka and Tzolis:
🗣️ “Different players like the ball in different ways. For example, as you said before, Bukayo, I know his quality. Sometimes he wants the ball early. And with Solly, he likes the ball early as well, or just more fired in behind sometimes because of his pace.
“But it’s important that I take pride in this stuff because a lot of our game play is passing, it’s connections between each other. So I pride myself on the fundamentals, and something that’s important to me is recognizing the right time to break lines, to progress the play, which is also a skill within itself.
“And what I’m learning and what I continue to learn is the right time to use different types of passing, which for me is super important as well.” #afc
Emily Whitfield retweeted
What Is #AI Agent Memory? Short-Term, Long-Term, Episodic & Semantic Memory Explained
buff.ly/PveS50e @UniteAi
Cc @jblefevre60 @sallyeaves @AkwyZ @ahier @rvp @gvalan @HaroldSinnott