
Biologist turned programmer. I like writing code and words. Code: @actuallymentor Words: https://nitter.gq/t.co/zVPWGZmgSz Current focus: Bittensor @TPN_Labs, @taofuxyz.
Amsterdam, The Netherlands
Joined May 2011
- Tweets2.1K
- Following1K
- Followers2.4K
- Likes8.5K
Pinned Tweet
I'm torn about whether I should pour money into SO2 balloons.
Also curious what's stopping a Musk, Besos, or even a @beffjezos from raising some money and saying "lfg, try to stop me".
Global warming is accelerating
There's only one thing we can do today to delay it before we burn, enough to solve the pbm: SO2 injection
Some ppl are squeamish about it but they shouldn't be. SO2 is so obviously the right solution that we should do it now. Here's why:
1
12
2,489
Mentor (actually) retweeted
We need to stop regarding political opinions as sitting on a one-dimensional “left/right” continuum. You may despise Donald Trump (as I do) while agreeing with him on one issue (as I do). Each one of us is better seen as a point in MULTI-dimensional space. The number of dimensions is open to discussion, as is their interaction with each other, and the feasibility of reducing their number because of correlations. Moreover, we need to form our opinions on each issue separately, in its own right, instead of identifying with a “tribe”, especially a tribe with its own social-media “echo chamber.”
I can barely express how disappointed in @LGE_Global I am.
I've been trying to get a developer account for my TV and it just doesn't fucking work.
- acc creation redirects to GB, I am NL -> err
- google login on web, tv lacks it -> err
- email acc creation, activation doesn't arrive -> err
I have so little faith in the stability of this company's infrastructure.
1
139
Mentor (actually) retweeted
NY Times: U.S. intelligence indicates that Jensen Huang had an affair with a spy.
U.S. intelligence: She is not a spy. We do not know what you are talking about.
NY Times: Well, what is a spy really anyway?
Scarcity can only be allocated three ways: price, queue, or lottery.
Pick your poison.
Hey @Substack, please don't make "recovery questions" mandatory to enable 2FA.
2FA = secure
Recovery questions = security hole
Why would you think information like my mother's maiden name is secure.
Do not gate 2FA behind a mandatory security hazard.
1
119
Please please please @OpenAI don't retire custom GPTs.
Plugins are NOT the same as custom GPTs.
Your in-app live voice is the best there is right now. Being able to "call" different personalities is a really useful thing.
I call my language tutor. I call my assistant. Please.
1
94
Always a pleasure talking AI and Bittensor nerdery!
Thanks for having me Mark
Hash Rate - Ep. 191: TPN Subnet 65
🧙 Guest: Mentor of @tpn_labs @Gratis_AI trueperformancenetwork.com
Try it! ---> gratis-ai.com
00:00 What TPN 65 does: compressing LLMs
08:27 Quantization: 16-bit to 8-bit
13:26 Fluid Bench and model-distribution scoring
17:16 Why Subnet?
21:29 Global competition, stranded talent
24:30 Open models, closed labs
26:03 Quantization-as-a-service
31:15 Mentor’s background and TPN’s pivots
40:08 Rapid protocol change versus long-term stability
47:18 Upcoming model releases, benchmarks, and demo apps
Hey @bunq will you please make year statements available on web, and not just the mobile app?
1
500
Als @odidonederland zou kunnen opschieten met mijn huis op glasvezel aansluiten zou dat fijn zijn.
@ZiggoWebcare is een groot storingsfestijn.
4
2
1,774
Oh my god I just tried to link an MCP to the @ChatGPT app for the first time.
This UX is completely unacceptable.
- Please don't make plugin creation be chat only, I have no idea what questions/inputs I can even ask it
- Authenticating is a maze to figure out
The fuk
1
91
First time user feedback for @posthog: your onboarding wizard is WAY too long and completely unskippable.
I just wanted to create an account and get the project/api keys I need.
The amount of "skip" and random "select things" boxes made the whole process frustrating.
1
1
100
The @AnthropicAI/@claudeai cyber filter is so frustratingly out of control.
Asking it to help secure something makes it think I'm trying to hack the pentagon or something.
This BS causes devs to either:
- not work on security
- use unhinged abliterated models
Either way, dumb
2
88
Honestly given the balance of effort/effect/incentives:
- 50% it's a teenager with Claude
- 50% it's a govt that needs the fear thermometer of the populace higher
Low chance it's an actual agressor. It would be the dumbest attack ever.
Attacking non-critical consumer-facing infrastructure - like grocery stores - with pre-warned and limited scope cyber attacks is the best way to convince your population they need to harden their society.
It's really classic Ender's Game - you only get to use the weapon once, it has to destroy the enemy completely.
Real catastrophe is not spoiling meat, its multi-megawatt turbines that go beyond safe RPM, critical valves in chemical refining that report the wrong position, air traffic control networks.
Destruction of hard infrastructure that is replaceable on a year-to-decade timescale. Like a water treatment plant, transformer station, airport, and so on.
Food is the most high throughput thing in our society besides gasoline. It has a residency time of a few days on shelves, and so any total loss of inventory is recoverable on the scale of a week.
Mentor (actually) retweeted
Introducing gratis-ai.com
Run open source AI models locally in your browser, in just a few clicks.
Pick an opensource model or load any model from @huggingface. It downloads to your machine and runs there privately; every prompt stays on your device.
Get your free personal AI today.
Mentor (actually) retweeted
We're excited to introduce @Gratis_AI.
Run opensource AI models locally on your machine for free, with no setup required.
Gratis runs models from @huggingface directly in a browser tab or desktop app for more power, making every chat private by design.
TPN builds small, capable models; gratis runs them.
gratis-ai.com
Introducing gratis-ai.com
Run open source AI models locally in your browser, in just a few clicks.
Pick an opensource model or load any model from @huggingface. It downloads to your machine and runs there privately; every prompt stays on your device.
Get your free personal AI today.
Hey @nickaturley is there talk of bringing Custom GPTs back?
Plugins lack live-audio and personality sandboxing.
E.g. With plugins I cannot "call" my Albanian language tutor anymore
The current chat-based plugin setup is also not my fave. It obfuscates configurable properties.
1
85
Ladies & Gentlemen, we have managed to get the @EUCouncil ' EU Inc files. The ones that matter. The ones where we see the position of Each Member state: including who pushed the notaries' agenda.
Buckle up and Repost. 🍿 Now if you are from those Member States, make some noise.
Directly or through associations, politicians. Enjoy the thread 🧵 1/
Mentor (actually) retweeted
What is TPN Labs building?
Join @ActuallyMentor and @VenturaLabs to delve deep into what TPN Labs is building through subnet 65.
Special thanks to Grant, the best host 🙏
$TAO
Ep. 106 - Mentor Palokaj
Mentor Palokaj @ActuallyMentor is building TPN @TPN_Labs
Timestamps
1:01 - From a TAO Bridge to a VPN Subnet
4:28 - The 440 Upgrade and a Six-Hour Rethink
6:42 - Profitable, but Not Parabolic
9:42 - What TPN Actually Does to a Model
10:03 - Quantization, Explained With a Ruler
11:36 - Quantization-Aware Distillation
12:25 - Nobody Knows How to Measure an LLM
13:15 - Anyone Who Can Cheat Will Cheat
13:49 - Fluid Bench, a Benchmark That Moves
15:13 - The TPN Workbench and Optimize
16:57 - An LLM as a Plinko Machine
20:02 - Toward Models That Are Experts
21:12 - DeepSeek on 8GB of VRAM?
21:30 - The Outlier Problem
25:18 - What Each Halving Really Costs
26:18 - Ternary Weights and Hadamard Rotations
32:09 - Is There an Optimal Bit Rate?
34:05 - Where TPN Goes Next
35:03 - Treadmills and a Teaser






















