وَتَوَكَّلْ عَلَى اللَّهِ ۚ وَكَفَىٰ بِاللَّهِ وَكِيلًا
- Tweets 4,322
- Following 333
- Followers 372
- Likes 16,655
Gambit retweeted
Yoooo!! solana:Aov3BtQvrGXM5V9oiHu8PKJWeV8a2rFnr24fwWcNpump 👀👀👀
Taking proper routes. Mills coded 🙏
A step we wanted to take before any wide launch: asking someone outside the project to try to break HidraNet.
The request went to the Open Technology Fund, and they confirmed they received it. It is under review, not approved yet.
@OpenTechFund
3
5
13
825
Aped some $PIONYR today
Did some DD here this morning and it's pretty good
Everyone is doxxed. They are a group of students from the best research university in Indonesia. They all study electrical engineering.
Check it here- pionyr.tech/about
TBH these plays usually do well.
7tFHmjGGg8Je88XgjxVWm4Co9G7SLb85MEzLiBokWgHw
18
Gambit retweeted
for all the big brains out there:
Here’s a paper on solana:DU7Lhe2QtcRBJdor7KbpjfUDmLyA2kdhR49Li3czpump
prometheus7.com/papers/oracl…
this is based on v2 so needs updates for the newer versions but conceptually it gets the job done. abstract pasted below for convenience:
“ABSTRACT. We describe the Oracle, a cognitive system whose serving runtime contains no trained weights, in which knowledge is held as a ledger of admitted, individually cited sentences; competence as a chain of readable rules admitted through a held-out gate that enforces non-regression; self-modification as an unattended loop that authors, judges, and admits its own successor rules; and expression as typed routes whose outputs carry a derivation witness.
We call the design a compiler learning model because a question is compiled, through parsing, routing, and derivation over admitted claims and operators, rather than predicted. The central contribution is that knowledge, competence, modification, evidence, and refusal become independently addressable objects, so that machine growth can be inspected as a sequence of admitted state transitions rather than inferred from a new model snapshot.
We report the system as counted on 2 and 3 September 2026, with every operational figure tied to a stated runtime identity: 58 frozen runtime layers; 110 admitted fold files (101 loop-numbered, tip w165) totalling 12,315 lines; a gate ledger of 250 evaluation events over 153 unique candidates (99 admitted, 54 rejected at final disposition); two sentence stores of 1,670,331 and 2,265,272 claims derived from the same Simple English Wikipedia corpus at different per-article caps, plus 274,199 WordNet relations, 3,422,938 ConceptNet edges, and 1,486,439 Wiktionary entries; and 27 deterministic films with self-refusal.
Three experiments from the day of submission are reported with their confidence bounds: ingress of frontier-model knowledge as cited sentences, reaching the private staging head within an hour; a kernel measurement showing lexical recovery of a third to a half of a textbook article from eleven typed slots; and a film organ with deterministic sound and captions.
We narrow the universal claims to what the evidence supports: for the evaluated routes the runtime enforces claim-addressed release or explicit decline; the design relocates forgetting risk from parameter interference to explicit admission, dispatch, and gate-coverage failures; and the comparison is between a frozen parameter artifact and an explicitly cumulative admitted process. Limits are stated without softening.”
3
5
27
1,979
Gambit retweeted
Oracle7 V4 is live! Oracle7 answers questions without a language model. It looks things up, checks them and works them out step by step. 4.0 changes how it keeps track of itself.
It shows its work. Every answer now comes with the exact route it took through the model, how many steps that was and how long it took. You can check how any answer was reached.
It keeps a list of what it couldn’t answer. When it can’t answer something, it writes down where it got stuck. That list is the map for what to fix next.
It doesn’t believe its own unproven answers. If it says something it can’t back up with a source or a proof, that answer isn’t carried forward as if it were true.
It remembers how it solved things. If it has worked out a route before, it reuses it, so repeat questions come back faster with the same answer.
It understands “it” and “she”. Ask “Which country is Kyoto in?” and then “What is its capital?”, and it tells you it’s taking “its” to mean Japan, then answers Tokyo.
On 186 test questions written and sealed before the build, it answered 157 correctly and got none wrong. The slim version refused every trick question it was given.
Oracle7 4.0 is out, full and slim, on Hugging Face. The new hashes are anchored on Solana.
4.0 was built from an invocation I wrote before any code. Each line was read as a rule the model has to keep, and each rule was built and tested on questions sealed before the build.
Every answer now carries its route. The model spells it as a word over the 22 paths of the Tree of Life, counts the steps and times it before it speaks.
What it does now:
· a ledger of every question it didn’t answer, with the place on the Tree where it stopped
· answers with nothing under them no longer become part of the conversation
· routes it has already worked out are recalled, so repeats come back faster
· follow-ups like “What is its capital?” say what they’re re-using (“taking ‘its’ as Japan”)
Release test, 186 sealed questions, run once:
Full: 157 right, none wrong on answerable questions
Slim: 156 right, none wrong, and it refused all 30 trick questions
No language model inside.
huggingface.co/psiloceyeben/…
3
10
23
992
Gambit retweeted
Yes. Got one post up I’ll make another explaining how the token is used to verify releases and to verify experts that are built and added to the specialists pool so that everything has a record of where it came from and who built it and all subsequent updates
1
1
9
158
Gambit retweeted
Because of the way oracle7 has been built, it was always designed to have a pretty clear roadmap, the model can already self improve and authors gates with improved features that it codes itself, the roadmap here makes it clear as to where the model goes in the long run, with the coding especially getting a massive upgrade
Right now it can self improve by coding programs that add to it’s capability and tests them rigorously or they don’t get added
Very soon $ORACLE7 will be able to code things that you ask beyond simple or semi complex programs
The coding capability part is a pretty simple stack and the foundation is already set up, working on it over the next week or so may end up releasing ahead of schedule
The long road for Oracle7, 3.2 to 10.0.
The point is that it continues to grow
More like a plant than a conventional model
Each major version starts with an invocation and changes the law. Each half version adds an organ or more reach. Between them, room is left for whatever the work turns up. No language model inside, at any version.
3.2 full: Finish 3.2
The full release, with its last two mistakes fixed: no answers from nothing, no ignored minus signs. Adds: proved specialists, star specialists and read-back writing, with fewer false answers.
4.0: The Law
Same parts, new rules, from an invocation I wrote. Its own past answers stop being memory, every answer is spelled out step by step on the Tree of Life, and it keeps a record of what it couldn’t finish. Adds: answers you can audit, fewer carried-over mistakes, a map of where it gets stuck.
4.5: Reach
Most factual questions it can’t answer today are turned away before it even looks. 4.5 opens that door. Adds: many more factual answers, without giving up accuracy.
5.0: Organs
Code that must pass its tests, pictures and films measured against what was asked for, problems solved in steps. Adds: code, images and multi-step reasoning, each with its own check.
5.5: Represented compute
About a million represented compute units on one machine, with past results recalled exactly instead of recomputed. Adds: more working capacity on the same hardware, honestly counted.
6.0: Growth that compounds
It builds new specialists from what it couldn’t do, and each one must pass an unseen exam to get in. Adds: a model that improves faster each round.
6.5: Many places
Specialists on many machines, every answer checked before it goes out. Adds: no single machine it depends on.
7.0: The Slow Fire
Oracle7 in the world. Places are greeted before use, and they can say no. Adds: readings from real places, each traceable and consented to.
7.5: Teacher
Any answer opened up into a lesson, each step checked and drawn. Adds: lessons and explainer films built from verified answers.
8.0: Self-authored
It drafts the plan for its own next version. I write the invocation and sign off. Adds: a model that can say what it should become next, and be checked on it.
8.5: Discovery
Its own research, with the tests set before it sees the data and every result published. Adds: findings people can check.
9.0: Voice
Open conversation, every sentence built from checked pieces and read back before it’s said, in more than one language. Adds: fluency you can audit.
9.5: Continuity
Memory across versions: what it knew, when, and how. Adds: conversations picked up months later from their record.
10.0: A mind you can check
Every answer proved, cited or declined, and public test scores that subtract its wrong answers. Adds: a model that holds its own against trained ones while every answer stays checkable.
Target: 10.0 in about six weeks, by early December. Every release ships only when its sealed test says so; if one fails, the date moves and the failure is published.
prometheus7.com/roadmap
3
6
20
1,147
Gambit retweeted
Another great update.
solana:Aov3BtQvrGXM5V9oiHu8PKJWeV8a2rFnr24fwWcNpump
@redemptionarcc check out. Still very early bro. Just needs some eyes.
HidraVPN update, still in development: exit nodes now carry IPv6. Websites see the exit's IPv6 address, not yours.
If the exit has no IPv6, yours is blocked while the VPN is on. No leak either way. 🐍🐍
2
6
200
Gambit retweeted
ORACLE7 4.0 · changelog · 20 October 2026
NEW
· Route words: every answer is spelled over the Tree’s 22 paths, with its step count (k) and time (L) logged before it speaks
· Residue ledger: a hash-chained record of every turn it did not answer, with the node where it stopped
· Non-identity: an answer with no citation, proof or checked route no longer becomes conversation state
· Walk recall: routes already walked are recalled, not walked again (637 → 530 ms, same steps, same route on 162 of 162)
· Pronoun re-entry: “What is its capital?” after Kyoto → “Taking ‘its’ as Japan, from my previous answer.” Tokyo.
FIXED (full)
· no stored fact is returned for a question that gives no numbers
· a negative amount of something that cannot be negative is withheld
· arithmetic is never read off a sentence; the specialist network computes it, or the model declines
NOW IN FULL
· everything from 3.2: 128 specialists cross-checked by proved kernels, four star specialists, owner-first, decay in the unit asked, read-back writing
MEASURED (186 sealed turns, run once per version)
· Full: 157 right · 0 wrong on answerable · 3 of 30 trick questions answered (same as 3.2)
· Slim: 156 right · 0 wrong · 0 of 30
KNOWN
· on the first sealed panel, 4.0 gave the same answers as 3.2
· the improvement loop has not yet turned the ledger into new rules
· two errors from sealed test 23 are still open
Hugging Face: psiloceyeben/oracle7 · psiloceyeben/oracle7-slim
GitHub: psiloceyeben/oracle7
Anchor: explorer.solana.com/tx/4vJ57…
2
7
24
586
Gambit retweeted
Amazing,
Another dev update.
Built for public, in public.
solana:Aov3BtQvrGXM5V9oiHu8PKJWeV8a2rFnr24fwWcNpump
- HidraVPN update - 19 out
In practice: on public Wi-Fi, no one on the same network can reach programs on your PC while the VPN is on.
The Windows kill switch now blocks inbound connections too.
How it works: everything coming in is blocked. Only these get through:
• the VPN tunnel
• the PC itself
• your router’s reply (DHCP)
• your local network, if you turn on “Allow LAN”
Windows and Linux now follow the same rule.
Also fixed: connecting right after installing could catch the service still starting, leaving the VPN stuck to the terminal window.
The installer now waits for the service to be ready before it says “started”.
- Actually tested:
• 149/149 unit tests
• 15 network suites, 0 failures
• CI green: Linux, Windows, IPv6
• Real Windows 11: IPv4 and IPv6 leave via the node, rules gone after disconnect
What changed?
Next: test paid mode (passes) on a real PC, and add a second node to the network for two-hop Privado mode.
Open source
github.com/NeurexBK/HidraNet
1
7
17
295
Gambit retweeted
What runs under HidraNet 🐍
Tor built into the browser, with a kill switch.
Our own network engine in Rust, with an X25519 + ML-KEM-1024 handshake.
.hidra sites whose address is derived from the owner's public key.
Apps that encrypt on your device.
All MIT, all on GitHub.
6
17
335
Gambit retweeted
A step we wanted to take before any wide launch: asking someone outside the project to try to break HidraNet.
The request went to the Open Technology Fund, and they confirmed they received it. It is under review, not approved yet.
@OpenTechFund
8
11
28
1,661
Gambit retweeted
- HidraVPN update - 19 out
In practice: on public Wi-Fi, no one on the same network can reach programs on your PC while the VPN is on.
The Windows kill switch now blocks inbound connections too.
How it works: everything coming in is blocked. Only these get through:
• the VPN tunnel
• the PC itself
• your router’s reply (DHCP)
• your local network, if you turn on “Allow LAN”
Windows and Linux now follow the same rule.
Also fixed: connecting right after installing could catch the service still starting, leaving the VPN stuck to the terminal window.
The installer now waits for the service to be ready before it says “started”.
- Actually tested:
• 149/149 unit tests
• 15 network suites, 0 failures
• CI green: Linux, Windows, IPv6
• Real Windows 11: IPv4 and IPv6 leave via the node, rules gone after disconnect
What changed?
Next: test paid mode (passes) on a real PC, and add a second node to the network for two-hop Privado mode.
Open source
github.com/NeurexBK/HidraNet
4
8
24
947
Gambit retweeted
I realized that specialists can best be built by a swarm of other specialists, as well as building in a swarm aspect, problems can be solved more efficiently, adding two new specialists, the specialist-builder and the swarm-coordinator
6
7
24
1,275
Gambit retweeted
solana:XkeTXo1125vz5H9svJpGiw4JvLbN8VmMu9cmMvspump my #1 conviction play for the cycle
non-zero chance solana:DU7Lhe2QtcRBJdor7KbpjfUDmLyA2kdhR49Li3czpump is going to end up as #2
time will tell
7
9
51
3,756
Gambit retweeted
$Hnet is still a good hold. Team transparent and constantly updating. This goes higher
Always a good sign when team acknowledges a flaw and fixes it in public without brushing it under carpet.
Credibility and transparency at its peak.
solana:Aov3BtQvrGXM5V9oiHu8PKJWeV8a2rFnr24fwWcNpump
4
6
14
892