Tech enthusiast, AI geek, and WoW gamer Exploring startups and snapping photos
Seattle
Joined May 2023
- Tweets2.2K
- Following270
- Followers18
- Likes5.3K
Samir Ghali retweeted
What's the most valuable phone call in your business? @illscience thinks AI will take that one first.
When @guilleflorvs and I asked Anish which spaces excite him most at @a16z, he went straight to voice, which he sees as the AI insertion point into the enterprise.
▫️ The first generation, last year, handled lower-value customer support calls
▫️ Now it's moving up to the most important calls an organization makes
His example is @happyrobot, voice AI for freight brokers. A trucker gets on the phone with the AI, negotiates the cost of a load, and ends up befriending it. In Anish's words, it's "essentially frontline sales."
His prediction: "the most important phone call that happens in a business is going to be handled by the AI and all the way down."
Building in voice? Reply with the call you're automating first.
xitter.cf/guilleflorvs/status/21…
BREAKING: After 1 year in stealth, @rdominguezibar and I are launching GR Angel.
We invest in exceptional founders, put them in front of one of the biggest tech audiences in the world: + 600k followers, + 400k subs & 10M weekly views.
Across our personal angel investments and GR Angel, we’ve now backed:
→ A portfolio that has raised $1.2B+
→ Companies including HappyRobot, Replit, Nozomio, Throxy, Apollo Atomics, Gru Space, Didit, Aktoria Robotics, Arga Labs, Hyperspell and QFEX
→ Alongside investors including a16z, Sequoia, Founders Fund and General Catalyst
→ 1 Exit: Portco went from $10Kmrr to 7M arr in close to 1 year
Our distribution has helped startups get millions of views, one doubled its ARR and another received a $1M investment offer and much more.
Now we’re looking for the most cracked founders worldwide, investing bigger checks and starting our growth phase, you can reach out to us 24 7 and hiring 👀
Samir Ghali retweeted
What startup would the CEO of @WeWork build today?
@guilleflorvs and I asked John Santora that in his NYC office for the Compound Podcast. He went contrarian.
Everybody is building AI in some way, so he looks at what all those companies need:
▫️ Power and data centers
▫️ Infrastructure and speed, so they can move more information quicker and faster
"I'd be on the kind of outskirts... because there's so many brilliant people and so much money in here."
The takeaway from that conversation: build the tools every AI company needs.
If you're building picks and shovels for AI, drop a reply with what you're working on. Guillermo and I read every one.
xitter.cf/guilleflorvs/status/21…
BREAKING: After 1 year in stealth, @rdominguezibar and I are launching GR Angel.
We invest in exceptional founders, put them in front of one of the biggest tech audiences in the world: + 600k followers, + 400k subs & 10M weekly views.
Across our personal angel investments and GR Angel, we’ve now backed:
→ A portfolio that has raised $1.2B+
→ Companies including HappyRobot, Replit, Nozomio, Throxy, Apollo Atomics, Gru Space, Didit, Aktoria Robotics, Arga Labs, Hyperspell and QFEX
→ Alongside investors including a16z, Sequoia, Founders Fund and General Catalyst
→ 1 Exit: Portco went from $10Kmrr to 7M arr in close to 1 year
Our distribution has helped startups get millions of views, one doubled its ARR and another received a $1M investment offer and much more.
Now we’re looking for the most cracked founders worldwide, investing bigger checks and starting our growth phase, you can reach out to us 24 7 and hiring 👀
Samir Ghali retweeted
ChatGPT might remember more about you than you realize
A circulating claim suggests AI memory systems can infer your personality, habits and personal details from everyday conversations, even when you never explicitly ask them to remember anything
The scary part isn’t just what AI remembers, it’s what it can figure out from everything you say
The alleged study and its 96% figure remain unverified
Samir Ghali retweeted
THIS IS PURE F*CKING INSANITY, CLAUDE JUST CRACKED AN NSA SECRET NOBODY COULD SOLVE SINCE 2000
It took under 30 minutes
In 2000 the NSA handed over the "random" seeds behind P-256, the curve your browser uses every single day
They never explained where those numbers came from
25 years of people whispering about a backdoor
A $36,864 bounty nobody could claim
Anthropic gave it to Claude with zero hints
- Spotted the top bit was 0 in every word → a 1990s random number generator
- Brute forced the hidden state in 131,072 tries
- Lattice reduction on the rest
- Ran the generator backward
- 160 out of 160 bits matched
The NSA's "random" seeds were the clock on somebody's computer between May 1997 and April 1998
No backdoor, nothing broken
Just a Unix timestamp that fooled the entire cryptography world for 25 years
If this sat in plain sight since 2000, what else is hiding in the standards we all trust?
Samir Ghali retweeted
CLAUDE CODE + YOUR TV: ONE PROMPT TOOK THE ADS OFF, 5,081 STARS IN 3 DAYS
You press Home on your TV and the first thing you see is a banner for a show you never asked about, then a row of apps you never installed, then a shop you will never open.
You paid for the screen.
The manufacturer rents out the front door.
A dev got tired of this on their parents' TV and tried something that sounds too simple: hand Claude Code the model number and let it work.
It found the TV on the wifi, backed up what was installed, switched off the manufacturer's junk and put a clean launcher in its place.
The home screen looks like an Apple TV now.
The video shows that run on a Xiaomi TV. On the left the session types itself out: connect to the TV, save the list of apps, switch off the bloat, make the new launcher your home.
In the middle scroll the real package names from the guide's Xiaomi lists: the analytics service, a music app, the update helper, the floating widget.
On the right the tiles go dark one by one with a line through them while the counter climbs.
Below it you get the three steps and what each one keeps.
The part that makes people trust it: nothing is uninstalled. Every package is only switched off, and one restore line brings the old home screen back.
That is why strangers dared to run it, and why copies for LG and Fire TV showed up within days.
The limits are honest too. Ads inside streaming apps stay, because they live in the apps.
On some Fire TV sticks the system refuses to switch things off without root.
And your package list will not match the one on screen, so let the guide discover your own.
If your TV's home screen annoys you, this is the cheapest fix you will find this week.
Samir Ghali retweeted
Building an agent skill is one thing. Knowing when it actually triggers is the harder part.
OpenCode Skill Creator is a workflow skill and TypeScript plugin for creating, evaluating, and improving OpenCode skills.
It helps you move from a draft to a tested skill by generating trigger evaluations, measuring description accuracy, and comparing task outputs before installation.
Key features:
• Skill drafting – creates a SKILL.md workflow with frontmatter and supporting files
• Trigger evaluation – builds should-trigger and should-not-trigger queries to measure skill selection
• Description optimization – improves descriptions using eval failures and a train/test split
• Run comparison – aggregates graded outputs, timing, and token usage across runs
• HTML review – lets you inspect paired outputs in a browser before installing a finished skill
It’s open-source (Apache License 2.0).
Link in the reply 👇
Samir Ghali retweeted
Had one of those nights where all the vibe coded pieces collapse in front of you and you realize engineers are still very, very much necessary
Samir Ghali retweeted
My prediction is that this will be true unless/until AI unlocks a *major* new way of exploiting the *physical* world, the way we got in the 19th and 20th century. Increased efficiency of the current paradigms are not gonna cut it.
Samir Ghali retweeted
Head of Claude Code:
"85% of our engineers are running dozens or hundreds of agents. The way you do it is graph engineering."
In 40 minutes he explains how a single engineer now does the work of a whole team, how far this has gone inside Anthropic and where it goes next.
These 40 minutes are worth more than any $500 agentic course you'll find online.
Watch it, then read the full guide on graph engineering below.
Samir Ghali retweeted
🔗 GitHub: github.com/truffle-ai/dexto
---
✉️ If you’re into AI, ML, agents, and building real systems, join my newsletter (it’s free): dankornas.substack.com
Samir Ghali retweeted
we had some fun news to share but got side tracked
we did our first test run on a new GPU rack with our new deepseek inference setup
the price/perf we're getting is better than every single vendor that we've found. and there's still more optimization to do
Samir Ghali retweeted
Reverse-engineering a binary shouldn’t mean living in a GUI.
Ghidra CLI is a Rust command-line tool for automating Ghidra reverse-engineering tasks, for analysts and builders who want scriptable binary analysis.
It helps you inspect and modify Ghidra projects from the terminal by connecting the CLI to a Java bridge running inside Ghidra, while keeping the project state resident between commands.
Key features:
• On-demand bridge – import and analyze commands automatically start the Ghidra bridge
• Function analysis – list, filter, decompile, disassemble, and update function signatures from the CLI
• Program exploration – query functions, symbols, types, strings, and cross-references without leaving the terminal
• Call graphs – generate caller/callee graphs and export them in DOT format
• Structured output – return results as human-readable text, JSON, or pretty JSON
It’s open-source (GNU General Public License v3.0 license).
Link in the reply 👇
Samir Ghali retweeted
A middleware engineer in South Korea wrote to me, seeking help on what to do now that their manager is "blinded by GenAI hype" and set an "AI token usage" KPI. This eng is using v little AI, so manager is poking them to use more.
My response: your manager is trying to help you
Samir Ghali retweeted
Im actively encouraging the members of the networking and network automation community get behind this movement and come out of your AI closet.
We will only figure this out if we figure it out together.
If the fear you feel is the judgement from the community and NOT the fear of using these new tools I'm here to support you. Lots of us are here to support you.
Samir Ghali retweeted
This is actually more a side-effect of the death of pretraining.
As transformers saturate, most pretraining results are tweaking data recipes, not massive stepwise changes in performance.
Turns focus to innovation in postraining and test time compute. This is good for competition, since pretraining concentrated resources in a few labs.
GLM 5.3’s biggest leap came from post-training alone.
“GLM 5.3 was built entirely by post-training the previous model. For me, that was a wake-up call.”
“Post-training is far more GPU-efficient than pre-training from scratch. Plenty of companies can’t afford to train a model from scratch, but they have proprietary data. They can post-train themselves and beat some of the closed models.”
“There’s still so much juice left to squeeze from post-training. We compared GLM 5.3’s cyber capabilities with 5.2, and it’s a night and day difference. That’s all just post-training.”
Samir Ghali retweeted
$MU $SNDK $WDC
DeepSeek is back at it again.
Diving deeper, this happened on a Saturday, no less, so the markets haven't had time to react. Only the weekend perps have.
The headline driving it is DeepSeek's claim that the new model needs 4x less memory per token.
That reads like less memory demand, until you look at the rest of the release:
+ The model is still huge. It has 552B parameters, and all of them have to sit in memory to run, even though only a fraction is active at a time.
+ It handles 1M tokens of context. Longer context means more memory per user.
+ It's priced at about $1.20 per million output tokens. Cheaper tokens have always meant more tokens used.
+ DeepSeek ran out of compute this week. It had a five-hour outage that experts blamed on not having enough capacity. A company that keeps making its models leaner and still can't keep its service up is evidence that demand is ahead of supply.
The weekend market has mostly recovered so far, so expect some reaction at Monday's open, but I wouldn't be surprised if the market shrugs it off by 10am.
Samir Ghali retweeted
GROK BOT SECOND BRAIN IS BECOMING F*CKING DANGEROUS
someone accidentally leaked this setup and it gives every trader a brain that remembers every call they ever made
20 minute setup ↓
01 ┃ make a folder called SECOND BRAIN
02 ┃ open a terminal in it, launch grok
03 ┃ gws CLI → Gmail, Calendar, Drive
04 ┃ create a MEMORY folder inside
05 ┃ AGENTS.md rule: every trade, call, mistake → MEMORY
06 ┃ dump your exchange history, trade journal, voice memos, chart bookmarks in
07 ┃ Graphify → it all becomes one knowledge graph
08 ┃ Playwright → it pulls charts and news on its own
09 ┃ /skillify → the whole thing is one command
then you ask
> why am i still not profitable?
↳ every big loss in your journal happened after midnight
↳ a voice memo where you called the move 2 weeks early
↳ 3 bookmarked setups you never took, all of them ran
"you find the winners, then bookmark them instead of taking them. stop trading after midnight"
it reads your trade history
remembers every call you ever made
and goes online to check what the market did after
your trading journal is a graveyard
this one talks back
how many winning trades are sitting in your bookmarks right now?
Samir Ghali retweeted
AI is making it more practical for devs to bring their game ideas to old consoles.
A developer used Claude Code to demake his Rust/Bevy auto-battler into Game Boy Advance and Nintendo 64 ROMs.
The full ports took roughly 2 days for GBA and 3 for N64, including party mode.
Claude could reuse the tested Rust core while adapting the graphics, controls and build process for each console.
A friend’s suggestion turned into two retro releases people can download and play.
For indie devs, that’s more room to turn unusual game ideas into playable ROMs.
Samir Ghali retweeted
For decades, infrastructure was expected to follow business strategy.
In the AI era, infrastructure may determine which strategies are possible.
Why is that still treated as a technical conversation rather than a board decision? #Strategy