Libertarian, AI/ML Specialist, EE, Atheist
United States
- Tweets 8,258
- Following 59
- Followers 277
- Likes 457
The stealth model Space Bunny Alpha has been mostly unmasked.
**It is NOT a model**, it's a router that uses GPT 6 and now GPT 6.1 SOL as main output model. It uses Deepseek 4 for hiding the real reasoning and Minimax tokenization is probably just a ruse.
spacebunnyalpha.com/#identit…
1
124
OpenAI has today burned Tibo @thsottiaux as a spokesperson.
They will need a new one soon as that clusterfuck of subscription devaluation is not recoverable by any other “gestures”.
Reset all you want, it’s done
1
17
If you label "Qwen Image" one more time "Open source model" a Tsumani may hit you
Thanks @arena for the recognition! 🏆 Qwen-Image-2.1 is now the #1 open-source model in both the Image Edit and Text-to-Image Arenas. Try it now and show us what you create! 🎨
2
94
Delusional…
Astra is 32 tk/second slow
It is the most inefficient choice openAI ever offered, costing subscribers so much that even 20x subs are emptied in a day.
And it has degraded since launch, heavily
2
62
How to deal with GPT Astra usage on Pro 20x accounts:
1) increase context size to 400k+ (it's free on Astra)
2) Use GPT Terra Max (or xhigh)
3) Now you use Terra Max to search for code, plan the implementation and then you ask it to send it to an Astra high or xhigh subagent. Ask to include code locations.
4) You can continue working with Terra while Astra works, let him plan the next part, send it to a second Astra. And so on.
- Tell Terra to "do not fork conversation history to Astra"
This gives you context control of Astra, which is exorbitant expensive if you let the session grow - so the cost now is in Terra - it's expensive enough like that.
Astra does all the implementation, Terra is there to give a plan to Astra which then can refine it while working.
Once the 400k context are being approached, see to start a new session. Best to leave the old Astra threads behind.
1
1
106
Has GPT Astra degraded since release ?
I ran the "Ana de Armas" portrait as SVG on GPT Astra x-high - comparing it with Astra from release day and the Qwen3.8-27B open source model.
There is a degradation but it's subtle:
* The hairline was flawless, and is now flawed a little.
* The eyes were well proportioned but now look a little less realistic.
* The face was awesome, now it is elongated.
* The earrings were perfectly placed, now they are a bit off.
* Release-Astra played with shadows, Now-Astra doesn't
Astra today looks to me like a high quality quantization, or kv-quantization. It's very close to before, just not really there - and always worse, not better.
Qwen 3.8 is still amazing, it beats every other model except for Astra and Fable 5 in this test. And Qwen is not that far away from them.
I tested this with other models, including Gemini, SOL and more - none of them are even close to those 3.
Photo included only as reference, the AI models did not see the photo.
3
423
Flow retweeted
WELL-ORCHESTRATED OP: "The truth is this was a well-orchestrated media op. to scare the public about AI in order to implement their preferred approach to AI, which is a government takeover, to have a very heavy hand in government regulation of AI." @DavidSacks
Chairman of the President’s Council of Advisers on Science and Technology, joins @BretBaier to examine the viral resignation of former Anthropic researcher Jacob Coxen. Sacks pushes back against the narrative of an organic phenomenon, arguing that the social media rollout and media coordination point to a calculated effort to stoke public fear.
101
483
2,195
96,707
Orchestrated like a product launch
00:02:35 Peter Wildeford shares WSJ ->
00:04:29 Jacob Coxon posts ->
+6m Nathan Calvin (Encode AI) amplifies ->
+10m Daniel Kokotajlo (AI Futures Project) amplifies.
Encode AI, AI Futures Project, and AI Policy Institute received Jaan Tallinn-linked funding
Tallinn also led Anthropic's $124m Series A
Obviously a paid marketing campaign behind this WSL "exclusive"
"We know how to control nuclear weapons... We don't yet know how to control AI."
Former Anthropic researcher Jacob Coxon, who recently left the company, is sounding the alarm on artificial intelligence, calling it "possibly the most dangerous technology that humanity has ever created."
Coxon, who says he spent three years conducting research at Anthropic and OpenAI, wrote in a viral post on X that the companies are more focused on beating each other to build the most advanced AI models than they are on safety.
He says an international AI arms race would be "disastrous," arguing global cooperation is the only way to manage the rapidly advancing technology.
@SpecialReport
2
131
This is the most uninformed post about anything related in technology I've seen since the days of Covid made every 2nd waitress an expert in protein synthesis.
We witness the same here.
If he was an important person at Meta model research, they have no future.
If OpenAI wanted to cripple an entire nation, they easily could today. All they’d have to do is remove alignment and unleash an agent swarm. It could probably within a day or so get access to all of the nations data centers and shut off all the country’s utilities.
Like, we are already past the point where AI can destroy the world. Do people realize this?
1
30
After spending about 60% of my weekly 200$ Pro allowance on GPT-6 Astra, I think it is the first model since the GPT-3 Instruct era (text-davinci-003) that feels like a genuine large step up.
Bugs that GPT-5.6 SOL Max failed to find after repeated attempts were found and fixed by Astra x-high in under 5 minutes.
Its speed has degraded noticeably over the last 24 hours, but the model itself is extremely good.
That's a much needed win for OpenAI
Kimi K4 and Qwen 4 have a high goalpost to tackle.
2
74
I have chatGPT Plus business seats and they fail after 2-3 simple short prompts, then Sol high usage is already spent.
OpenAI models are NOT known to be efficient, it’s the exact opposite.
The 240$ subscription can do 10-12 minutes of inference per 5 hours, total 1 hour a week.
I've enjoyed working with Cursor pre-acquisition and have respect for the team and what they have built. The 5% here should have come with strong caveat and I would love for Michael to share the math.
Tokens are not a proxy for revenue nor value created and the OpenAI models are on the very frontier of token efficiency. Smaller or less strong models require many more tokens to achieve a task and therefore will inflate traffic share significantly.
1
47
This is a misleading graph, and totally false.
OpenAI is #1 customer of Microsoft Azure GPU use and Github Copilot was heavily based on GPT 5.5 up until June.
Copilot essentially closed business on 1st June, the majority of developers and coders using it switched to Claude and to OpenAI Codex.
That's the "user growth" - when in reality they LOST users to Claude.
87
These tests are done in „Max“ mode, nobody uses max for economic reasons. If you do the test in „High“ reasoning then Qwen beats GPt 5.6 Sol !
And the recommended reasoning is „Medium“.
The reality is crazier than even this test shows
36
June 2: White House signs EO “Promoting Advanced Artificial Intelligence Innovation and Security.”
It explicitly says the government will not create mandatory licensing, preclearance, or permitting requirements for the development, publication, release, or distribution of new AI models.
June 9: Anthropic releases Claude Fable 5 — one of the most capable frontier models ever made public.
June 12: The U.S. government issues an export control directive forcing Anthropic to disable access to Fable 5 worldwide (and Mythos 5) for everyone, citing national security.
The EO’s promise lasted 10 days.
whitehouse.gov/presidential-…
1
683
This guy appears to be involved in directly violating the Executive Order from June 2nd.
whitehouse.gov/presidential-…
An ignorant, short sighted and uneducated violation of section 3c less than 2 weeks after it was signed by Trump:
"Nothing in this section shall be construed to authorize the creation of a mandatory governmental licensing, preclearance, or permitting requirement for the development, publication, release, or distribution of new AI models, including frontier models."
I’ve had a number of conversations with folks inside and outside government about the current situation with Anthropic, and here is what I believe to be true:
— As we know, Anthropic publicly released its Mythos class models earlier this week under the commercial name Fable.
— Fable is Mythos with guardrails. But if those guardrails fail, then you’ve exposed Mythos and its advanced cyber capabilities to people who shouldn’t have them. (Keep in mind that Anthropic itself widely promoted the idea that Mythos was a cyberweapon and needed to be regulated as such. They asked for government regulation of Mythos and championed the guardrails on Fable. If there is a vulnerability — big or small — it is Anthropic’s responsibility to patch.)
— A highly credible trusted partner of both Anthropic and the USG who was testing Fable came forward with a jailbreak of those guardrails. The Admin asked Dario to fix the jailbreak or de-deploy the model. Dario refused.
— In their blog post, Anthropic defended its decision by saying the jailbreak isn’t serious. That is not what the trusted partner and the USG believe; nor is that kind of minimizing language consistent with Anthropic’s brand as the AI safety company. It’s difficult to fathom how they could claim a jailbreak allowing operability of a cyber weapon could be defined as not “serious.”
— In the past, Anthropic has always said that safety must be top priority and taken super seriously. In this case, Anthropic prioritized the continued offering of the consumer model over safety.
— In reaction, the Admin issued the export control. The Admin did this reluctantly. It’s been very surprised that Anthropic hasn’t wanted to cooperate with a reasonable safety request (ie fixing the jailbreak issue). Anthropic’s reaction is very much at odds with their branding and ethos as a safe AI research community.
— The Admin’s hope now is that Anthropic remediates the safety issue, the export control is lifted, and Fable goes back into general release. The Admin wants all of this to happen as soon as possible. It is frankly bewildered that Anthropic hasn’t wanted to comply with safety requests that it previously said were its highest priority.
— Those trying to misdirect and tie this action to the prior DoW/Anthropic issues are wrong. The Admin values Anthropic’s technical capabilities and feels that this issue, while serious, should be easily resolved. The ball is in Anthropic’s court.
1
35
Deep comparison of all Agentic Providers
Github Copilot increased prices from June on, hiking their 39$ subscription to up to 30,000 USD monthly!
So what are the alternatives, can we reduce spendings, is using opensource an option, local models, how does Claude, Codex and Cursor compare to Copilot today ?
What is the cost ?
Let's assume 1 hour of inference on a good model (like GPT 5.5) per day for 30 days - on an average codebase.
Copilot cost: 4000$-8000$ per month
Codex cost: 100$-200$ per month
Claude cost: 200$ per month
Cursor cost: 500$-2000$ per month
The Codex cost is fluctuating between 100 and 200 USD, it depends on the random regular limit resets.
Claude comes with stable 200$ a month, Cursor is expensive for frontier models and will cost up to 2000$ if you insist on good model quality.
Copilot costs 4000$-8000$ - in June around 4000$ but the "flex credits" are no promised to last, so it can easily double to 8,000$ next month.
The copilot cost MIGHT increase by another 30% due to inefficiencies in the Copilot harness, it is designed for a flatrate and not for conservation of tokens.
BYOK alternatives
Thankfully, if you log of Github completely you'll suddenly change the Copilot chat extension into a "BYOK" mode.
GLM 5.1 is around Sonnet level, below GPT 5.5 reference. Deepseek V4 Pro is also strong.
What is the cost?
GLM 5.1 Plan: 15$-50$ per month
Deepseek Pro: 200$ per month
Winners and losers
Loser Overall: Copilot, 300+ times more expensive
Winner Budget: GLM 5.1 - Sonnet quality for 30$
Winner Premium: Codex - best for complex work
Second Place: Claude - best for frontend work
Middle runner: Cursor - only if you use Composer 2
3
8
2,602
Europe is going directly for a military confrontation against Russia. And the US does not intend to be forced into it when the Russian answer kills their stationed troops.
This also marks a set of dissolving NATO - article 5 was voluntary but the backlash of ignoring it would be large.
The US is unlikely to enter the European war and make it into a world war. China likely also stays out of it
1
5
353
