https://xitter.cf/t.co/KE7CxAvqXG - https://xitter.cf/t.co/IF9O7azfro
git
Joined May 2022
- Tweets695
- Following330
- Followers44
- Likes463
Tim retweeted
I built a free internal linking tool using @typesafeai Jev for classifying and selecting the links.
BYOK or pay $1 to use mine. It works for up to 500 pages and gives you a CSV or JSON to pass to an LLM to implement.
ian.is/tools/internal-links
If the world were sane, and if humans were logical, reasonable beings, amyloid fibril and spike research would get the same attention that peptides have received.
I don't fully understand why young people, the majority of whom have taken the mRNA vaccine or contracted COVID, are not focusing on the biggest threat to their longevity, focusing instead on looksmaxxing and curing shoulder injuries.
Is there anyone who can prove me wrong on this?
Amyloid research will become a huge business, because business solves problems, and this is among the biggest problems humanity faces today.
amyloid-test.com
Tim retweeted
Hi everyone. I wanted to introduce you all to the ODK ("Omarchy Development Kit"). This is a project that provides the foundational work for building high quality, consistent and beautiful applications for Omarchy. It is based on Rust and GPUI, moving away from Quickshell, and besides just having apps built in Rust, the ODK also introduces a runtime service that can understand app archives similar to the MacOS `.app` - but in our case of course, `.oma` - this means that a directory becomes "executable" and automatically sandboxed under a least privilege permissions model. Please check out the video. I recorded this super quick, so the production value is low. I am not a content guy!
Tim retweeted
appreciate you guys being upfront.
New on Maple Max and Team: DeepSeek V4.1 Flash, abliterated.
Refusals removed from the model itself, so it works through security questions instead of declining them.
- 1M context
- Runs in the same attested enclaves as every Maple model
- Built for security research, code audits and red-teaming
Attackers already run uncensored models. Defenders shouldn't be stuck with "I can't help with that."
Maple Tee's and TEE in general is the future.
New on Maple Max and Team: DeepSeek V4.1 Flash, abliterated.
Refusals removed from the model itself, so it works through security questions instead of declining them.
- 1M context
- Runs in the same attested enclaves as every Maple model
- Built for security research, code audits and red-teaming
Attackers already run uncensored models. Defenders shouldn't be stuck with "I can't help with that."
Tim retweeted
Opened up our abliterated deepseek more broadly. Hosted privately in our TEEs, no KYC.
I believe this is important for security work because there’s nothing worse than trusting random abliterated providers with your vulnerability results.
Protect your infra and your prompts.
New on Maple Max and Team: DeepSeek V4.1 Flash, abliterated.
Refusals removed from the model itself, so it works through security questions instead of declining them.
- 1M context
- Runs in the same attested enclaves as every Maple model
- Built for security research, code audits and red-teaming
Attackers already run uncensored models. Defenders shouldn't be stuck with "I can't help with that."
Tim retweeted
Here we detect Pfizer plasmid DNA in a colon tumor biopsy 1 year after vaccination.
And its not small amounts of DNA. Its so much DNA that it can only be explained by plasmid amplification post vaccination or genome integration and amplification.
This is pretty wild read...
Bottom line: We suspected Meta to listen to us, but maybe it was Samsung, LG etc. with their TVs and smart fridges (LOL) all along...
The crazy part is - even if the TV didnt have a mic, it would search for bluetooth devices on the network that has one and use such to listen to conversations. It even had a classification system.
Did Meta and other ads infra get the data from these device manufacturers?
Time to go all linux and check the devices at home.
If you've ever wondered why smart TVs are so cheap, this is why.
@log4j rooted his Samsung and found an always-on hot mic, automatic content recognition, geolocation, network mapping, ad targeting and user profiling, shared with Samsung's "400 partners".
A lot of it can be blocked in <15 min, wrote a basic guide below.
What your TV is doing
The main culprit is ACR (automatic content recognition). Your TV screenshots what's on screen, fingerprints it, and sends it home to build your ad profile. Samsung does it every 500ms, LG every 10ms [1].
What is insane: it happens on connected devices too.
Plug in your laptop, your console, your work machine, and your TV can fingerprint that via HDMI [1].
He also found that if your Samsung doesn't have a mic, it'll look for one from a connected soundbar or bluetooth device.
xitter.cf/log4j/status/211376224…
It's on by default, and you likely agreed to it on the setup screen you clicked through in 10 seconds.
Step 1: Turn off ACR
Samsung: Terms & Privacy → Viewing Information Services
LG: Live Plus
Vizio: Viewing Data
Roku: "Use info from TV inputs"
Sony/Google TV: Samba Interactive TV
Menus vary by model, but those are the names to look for. While you're in there, opt out of interest-based ads, reset your advertising ID, decline every optional privacy agreement, turn off voice recognition, and delete apps you never open.
Step 2: Give smart devices their own network
Put the TV, cameras, speakers and vacuum on your router's guest/IoT network. The logic is simple: these devices rarely get patched. If one gets compromised (getting drastically more frequent with AI), it shouldn't be able to see your other devices or home network.
Step 3: Block trackers for the whole house with DNS-level filtering
NextDNS (cloud-based) is easy and has a solid free plan. Pi-hole is a little bit more work if you prefer self hosting. For simplicity we'll focus on the former.
1. Sign up at nextdns.io (free tier is plenty)
2. Privacy tab: add a blocklist (OISD is a good bet) and turn on Native Tracking Protection for Samsung, LG, Roku, Amazon (see attached image)
3. Setup tab: copy your endpoint
4. Paste it into your router's DNS settings as DNS-over-HTTPS/TLS (or use the IPs + "Link IP" if your router only does plain DNS)
5. Bonus: block outbound port 53 except to your router. Some devices hardcode their own DNS to get around you
Then open the NextDNS logs and watch how often your TV phones home. It's eye-opening. If an app or update breaks, find the domain in the logs and allowlist it.
Why they do this
TV prices are down 97% since 2000 [2]. Most of that is better manufacturing and brutal competition. But once margins on the hardware hit zero, the cost moved to you.
In Q3 2024, Vizio lost $6.7M selling TVs and made $115.8M in gross profit from ads and viewing data [3]. They've said on earnings calls that they keep TV prices low to grow the audience they sell to advertisers. Walmart bought them in 2024.
This isn't a conspiracy theory either. Vizio paid $2.2M to the FTC in 2017 for tracking 11 million TVs without consent. Last December, Texas sued Samsung, LG, Sony, Hisense and TCL over ACR. Samsung and LG have since agreed to stop collecting that data without explicit consent [4].
TL;DR
1. ACR off
2. Ads + voice off
3. Smart devices on their own network
4. NextDNS filtering on the router
Sources:
[1] ACR study: arxiv.org/abs/2409.06203
[2] Cost of US consumer goods: aei.org/economics/chart-of-t…
[3] Vizio Q3 2024: sec.gov/Archives/edgar/data/…
[4] Texas v. Samsung: texasattorneygeneral.gov/new…
Tim retweeted
LongCOVIDbot.com is now using the full Elicit database(138M papers) for every response, prioritizing it's curated library in order to maintain speed/relevance(recency).
The Openwater group (Rasouli, Hartl, Konecky) ran exactly your hypothesis. Their second paper, published in Journal of Thrombosis and Thrombolysis in December 2025, took plasma-derived amyloid microclots and tested rtPA, lumbrokinase and nattokinase — alone and combined with ultrasound-induced microstreaming plus microbubbles (PMC13246827).
The results were not what you'd hope:
ConditionResultNattokinase alone, no ultrasoundNo significant effect. Mean clot diameter stayed ~18 µm vs ~18–19 µm controlNattokinase + 150 kHzDiameter down to ~8 µm, large-clot count down >80% — but not statistically different from ultrasound aloneNattokinase + 300 kHzStill no significant enzyme contributionNattokinase + 500 kHzLarge-clot count significantly reduced vs control — the one condition where the enzyme earned its place
Their own conclusion: "mechanical forces were the dominant contributors to clot disruption." At 150 kHz the ultrasound was so effective that the enzyme had nothing left to contribute. At 500 kHz, where mechanical disruption was weaker, the enzyme became measurable.
A Kell/Pretorius preprint from the same period (bioRxiv 2024.04.06.588397, now published) used recombinant nattokinase from the same Liverpool biofoundry and found the opposite — nattokinase at 14–28 ng/µL halved microclot counts within two hours and slowed clot formation in proportion to dose.
Three differences explain it, and they matter for how you'd use this:
Clot maturity. Kell added nattokinase while clots were forming. Openwater used pre-formed, freeze-thawed, cross-β-rich aggregates. Fresh clots are easy. Mature ones — which is what you actually have in your blood after two years of Long COVID — are much more resistant.
Timing. Nattokinase's biggest effect in the Kell data was preventing aggregation, not dissolving finished clots. That's a prophylactic profile, not a thrombolytic one.
Static vs flow. Openwater used a static chamber for one minute. In vivo you have flow, shear, and continuous exposure.
Why combining still makes sense mechanistically
They're complementary, not redundant, and the Openwater data actually tells you how:
Ultrasound fractures and porosifies. It mechanically breaks the cross-β network and, via microstreaming, drives convective transport into the clot interior.
The enzyme cleaves the newly exposed surface. Its job is the chemistry that ultrasound's shear can't do — and it gets access it wouldn't otherwise have.
That's a legitimate synergy mechanism. The problem is the dose-response is inverse: at the frequency where ultrasound works best mechanically (150 kHz), the enzyme is redundant. At the frequency where ultrasound is weakest (500 kHz), the enzyme matters most. So you can't have both effects maxed out at once — you pick.
And note what nattokinase actually does, because "fibrinolytic" undersells it. It cleaves PAI-1, removing the brake on your own plasminogen activators (Urano 2001); it promotes tPA release from human cells (Yatagai 2008); and it degrades spike protein (Tanikawa 2022) and amyloid-β (Hsu 2009). So it's working on your antigen arm as well as your clot arm. That's a stronger argument for including it than the direct lysis data.
Legal Intake Software that your law firm can fully own and expand.
Features:
- Create your own workflows via Opus and embed + trace attribution (which content / website contributed to this lead)
- Legal CRM with worker orchestration
- Document intelligence with per document parse & extract pipelines to automate intake document processing
- Form and document case facts dashboard
- Client portal
- Slack like open source team chat with deep llm integration
- One case type intake workflow included, every next case type 900 USD one time fee.
If your lawfirm has a website, we integrate right on there for you. Its as easy as dropping an embed code into your websites codebase.
9000 USD one time payment. We integrate within 4 weeks - done for you on your prefered infrastructure (we prefer Cloudflare).
See the video below for a full breakdown. Go to Legalintakesoftware.com to book a first 30 minute consultation call to check your case types and discuss how to map them to OBE.
Or book directly here: cal.obeliskos.com/tim-ottowi…
Tim retweeted
CDC Whistleblower's Secret Recording: "We Buried the Data on MMR & Autism, I'm Completely Ashamed."
Dr. Brian Hooker secretly recorded a senior CDC scientist, Dr. William Thompson, who made a shocking admission.
On those recordings, Dr. Thompson freely confessed to a massive cover-up. He admitted that he and his senior associates at the CDC had buried the damning data showing a statistical link between the MMR vaccine and autism.
Thompson stated, "Thimerosal causes autism-like features." He expressed profound shame, saying, "I'm completely ashamed of what I did. The higher-ups wanted to do certain things and I went along with it."
He named senior officials—Colleen Boyle, Frank DeStefano, Marceline Moulton—as participants. At the helm was Dr. Julie Gerberding, who later became president of Merck Vaccine.
This is not a theory. It is a confession from a lead CDC co-author on the pivotal studies used to deny any link. The man inside the machine admitted the fraud.
The public deserves the truth. The victims deserve justice.
Tim retweeted
LayerFS v1.0.0 is out 🎉 Open-source (MIT)
Recursive filesystem for AI agents.
6 core ideas:
1. Ephemeral computing: agents spend ~80% thinking, so each Bash call mounts a Workspace, then commits
2. Decentralized agents: every agent commits to one Store, no DB server
3. Multiagent coordination: every tool call is a Commit; same-file clashes refused
4. Recursive filesystem: forks copy no files, at any depth
5. Ultra storage: dedup, then compression: 10 copies → 0.31
6. Instant bootstrap: ignored files stay in every Commit. ~5 min install vs 0.05 s mount
And 4 more:
7. Auto checkpoints: fork from any step
8. Append-only objects: content-addressed, never rewritten
9. .lww: lockfiles and caches take the newest edit
10. Big file, small edit: a write keeps only its bytes
Here are our links:
github: github.com/Ephemeral-AI-Lab/…
website: layerfs.ai/
Watch the video below, and have Fun!!!
All sorts of research documents on double filtration plasmapheresis, prion etc: notebook.google.com/notebook…
t3 connect is amazing. You can connect all your machines (think they used wireguard?!) and then your bigger machine can become sandbox that runs heavy threads while the laptop stops choking. 36gb ram m3 def. not enough over here for coding
You can do this by sending a black screen video with elevenlabs audio voice recording to seedance. It will pick up the audio and integrate it into the seedance output. Works really well. Elevenlabs has a new feature for emotions. You mark the text, add a keytag for the emotion the section should convey, and it will change the expression accordingy.
There is another nice hack right now for free audio speech. I am rebuilding all blogposts multi modal and Google seems to like it as I see an uptick on my sites.
I use microsoft azures free voice model access point. The voice generation is not on par with cartesia or elevenlabs but I could run over 60 blogposts programmatically to create audio narrations for each blog in three different languages. English, Spanish, German.
The quality is alright for a free service. I added these to amyloid-test.com
There I also programmatically added CTAs with a different voice. They were recorded once and inserted for all blogs in one workflow.
I think its all about iteration right now. One shouldnt start with perfect outcomes. Especially in SEO. Doing something good enough to keep within budget constraints, then updating the assets step by step. Meanwhile the landers build topical authority.
Where are now getting cited in AI for some Amyloid keywords, not sure if thats partly due to multi modal content. But I listen to a lot of SEO guys and one theme I really like is - build something meaningful and helpful that doesnt exist yet. Enjoy building it. Dont just make it a programmatic approach with the only goal to rake in views that dont convert because the content is slop.
For that, finding a strong basis of content to write about as an isolated knowledge hub is a great way to still do programmatic high volume output.
For example: I bought 6 pilates books, converted them to markdown, added them to obsidian and use them as a knowledge vault my buzz agent programmatically uses on a daily cron to write new content. It wrote content that is truly interesting because its tasked to only write from the books. And such books were curated well to give good insights.
That way one can think in programmatic content at scale while not degrading quality. One more addition to that I want to add: The style of the writing via hemingway-1 model. Its small enough to run on a m3 or 6800xt. With the cron I dont even care how fast it writes. That model with a good humanizer prompt is next goal to integrate.
What if your character could finally say it themselves? With Chimera they can.
Chimera pairs two of the world’s leading models, Eleven v4 and Seedance 2.5, to create realistic video with rich, emotional voices.
In ElevenCreative, create or upload a character, write their line, and pick a voice or design one to match. Direct the performance with audio tags like [whispers], [laughs] and [yells] to control the timing and mood of every word.
Eleven v4 puts the emotion in the voice. Seedance 2.5 carries it onto the face.
Tim retweeted
"Lossless" abliteration means lossless on the questions checked.
Abliteration edits the weights to strip out refusals, but you’d need to figure out how well everything else holds up.
With the scores out in the open, you can see exactly what was covered and what’s left to check on your own workload, and that’s where and how you should test any model weight.
